Security

Click to know more

Security Topic 1:

Automated Firewall CE Self-Assessment Tool Using Wazuh and pfSense

Research aim

This research aims to develop a practical solution by building an automated tool that simplifies the Firewall CE self-assessment (FCESA) process, so that it meets Cyber Essentials standards for firewall configurations and provides recommendations to improve security posture of the organization and reduce risks.

Research Question

To set up a virtual environment to simulate different network scenarios.
To integrate Wazuh and pfSense for collecting and analyzing firewall logs and configurations.
To develop automated scripts for assessing firewall configurations against CE standards.
To analyze collected data against CE standards using data analysis tools.
To generate comprehensive reports summarizing assessment results and providing actionable recommendations.

Research questions

RQ1: How effective is the automated tool in identifying and assessing firewall configurations compared to manual assessments?

Short Description of the idea

The purpose of the study is to automate the Firewall CE self-assessment process to enable a better compliance with Cyber Essentials standards. Traditionally, self-assessments and audits have been laboriously done by hand in a way that makes them susceptible to human error. This research builds an automated tool for data capture and analysis of firewall configurations aimed at the reduction of manual efforts during assessments. The project main innovations comprise the unification of the open-source tools Wazuh and pfSense for real-time monitoring, log collection, and compliance analysis. The benefit of automation will be that an organization can improve its security posture while achieving compliance more efficiently. Such an approach reduces the likelihood of human errors, thus bringing about more consistent and reliable results and enhancing the efficiency of security assessments.

Research Methodology

In this research, a quantitative approach will be taken, starting from a literature review followed by experimental analysis. The first phase will define the specific requirements for the Firewall CE self-assessment by referencing IASME’s Cyber Essentials question set. Then a virtual environment will be set up using VMware or VirtualBox, where virtual machines will be configured to simulate different network scenarios. Wazuh will be installed and configured to monitor the firewall logs. After that pfSense will be downloaded and installed on a bootable USB, with network interfaces and firewall rules according to security needs. Integrating Wazuh with pfSense will ensure the efficient collection of logs. Automated scripts in Python or Bash will be developed to assess firewall configurations against CE standards. Data collection tools will gather the necessary information, which will be analyzed using Pandas or Excel to generate reports and recommendations to improve firewall security.

Click to know more

Security Topic 2:

Automating Secure Configuration Self-Assessment for Cyber Essentials Compliance Using Open-Source Tools in a Virtualized Environment

Research aim

The study is aimed at the development of an automated self-assessment tool for Secure Configuration under the Cyber Essentials framework, hence reducing misconfigurations through automation, and increasing the efficiency and accuracy of self-assessment in comparison with the traditional method.

Research Objective

To set up a virtualized environment using VirtualBox to simulate Windows and Linux machines.
To identify main access parameters that are secured according to Cyber Essentials guidelines.
To develop scripts using Bash, PowerShell, or Python to automate system security checks.
To integrate open-source security tools (e.g., Lynis, OpenSCAP, OSQuery, Wazuh) for automated assessments.
To generate detailed security reports identifying misconfigurations and providing remediation steps.
To evaluate the effectiveness of automation by comparing results with manual assessments.

Research Questions

RQ1: How can automation improve the efficiency and accuracy of secure configuration self-assessments within the Cyber Essentials framework?
RQ2: As open-source tools (Lynis, OpenSCAP, OSQuery, Wazuh) compared to manual assessment, how effective are they in finding misconfigurations?

Short Description of the idea

This research paper is aimed at developing an automated self-assessment tool for Secure Configuration under the Cyber Essentials framework in the UK. Traditionally, organizations tend to run into problems that come with performing manual self-assessments since they may take longer to examine security misconfigurations due to inefficiencies and human errors. This method has turned out to be time-consuming and inconsistent, hence heightening the chance of non-compliance. This research, therefore, intends to enrich the self-assessment procedure by using open-source security tools and a virtualized environment to evaluate system configurations for Windows and Linux systems. By utilizing the technologies, Lynis, OpenSCAP, OSQuery, and Wazuh, the research seeks to develop an automated low-cost and scalable solution that supplements organizations in enhancing their cyber resilience and compliance efforts. In essence, this project works on the need for least-effort security assessments to be further augmented to assist organizations in mitigating vulnerabilities and achieving Cyber Essentials certification in a more accurate and timely manner.

Research Methodology

The quantitative methodology of this research will begin with a thorough literature review and experimental analysis. The first phase will deal with examining the Cyber Essentials guidelines regarding the fix requirement for Secure Configuration, focusing also on the existing security assessment tools such as CIS-CAT and OpenSCAP. The next step will involve establishing a virtual lab where Virtual Machines (VMs) will be deployed for Windows and Linux using VirtualBox, thus intentionally configuring these operating systems for the test through the introduction of security misconfigurations. The selection of the tools and their actual implementation are expected to employ Linux tools such as Lynis, OpenSCAP, and OSQuery in combination with various Windows tools such as PowerShell scripts, Wazuh (SIEM), and OSQuery. The automated assessments will be done by writing custom Bash and PowerShell scripts for these assessments. Security configuration scanning of the VMs will then take place to get information about misconfigurations, vulnerabilities, and compliance status, after which an analysis will be done by comparing the automated output against the output of a manual assessment, pin-pointing gaps encountered in detection and analysis algorithms, thereby generating reports and recommendations for future conduction of assessments.

Click to know more

Security Topic 3:

Automating User Access Control Self-Assessment for Cyber Essentials Compliance in a Virtualized Environment

Research aim

The aim of this research is to design an automated self-assessment tool for User Access Control (UAC) compliance with Cyber Essentials, targeted at supporting non-IT SME owners, through its ease of providing security assessments of access controls, lowering dependence on expert consultancy costs through effective automated guided remediation.

Research Objective

To design and develop a user-friendly tool that assesses user access control settings in Windows and Linux.
To enable non-IT SME owners to generate both standard and ad-hoc reports on their security status.
To provide comprehensive remediation guidance for resolving identified access control issues.
To implement continuous monitoring of user access controls to ensure ongoing compliance.
To utilize licensed and open-source tools that are reliable for security analysis.
To store security assessment results in a database (MySQL/Access/Oracle) for effective tracking of improvements.
To ensure compliance with Cyber Essentials guidelines through automated assessment checks.

Research Questions

RQ1: How can automation improve the efficiency of User Access Control self-assessments for non-IT SME owners?
RQ2: What is the relationship between exposure to violent video games and aggressive behavior in individuals, and how does this compare to individuals exposed to non-violent video games?
RQ3: How effective are open-source/licensed tools (OSQuery, Wazuh, Auditd) in detecting UAC vulnerabilities?

Short Description of the idea

The project primarily concentrates on the automation of User Access Control self-assessments to assist non-IT SME owners in obtaining Cyber Essentials compliance. This project aims at the development of an easy-to-use self-assessment tool that would ease the process of checking user account security, privilege levels, and access policies. Earlier, poor access control configuration plagued SME owners and thus depended too much on expensive consultants for evaluations. This project aims at solving these problems by providing a solution that would not only identify vulnerabilities but also provides comprehensive extensive step-by-step remediation. The research introduces a number of technologies and tools, including Java/Python in tool development, PowerShell and OSQuery for Windows security auditing, and Auditd and Lynis for Linux assessments. The tool, by embedding continuous monitoring and reporting capabilities, will ensure ongoing compliance and empower SME owners to effectively manage their cybersecurity posture, thereby reducing their dependence on cybersecurity consultants and increasing the safety of the digital environment.

Research Methodology

The research methodology for this study adopts a quantitative approach incorporating literature review and experimental analysis. Firstly, literature review will be conducted in order to find out the research gap. This will be done by exploring several databases such as Google Scholar, IEEE, ScienceDirect, and MDPI. Experimental analysis will be undertaken afterward, consisting of various developments over several phases. Phase 1 involves research and requirement analysis, focusing on the Cyber Essentials UAC requirements outlined in IASME documentation, while also understanding access control challenges faced by SMEs and evaluating both open-source and licensed tools suitable for automated UAC assessments. The second phase requires the design of a prototype tool in either Java or Python, along with the creation of a complete virtual laboratory with Windows and Linux virtual machines and the development of test user accounts with different privilege levels. Phase 3 sees the implementation of windows security auditing with PowerShell scripts and Wazuh for monitoring, while Linux security auditing utilize Auditd and Lynis. Finally, Phase 4 focuses on the assessment of tool accuracy against manual assessments, improved usability for non-technical users, and ensuring real-world applicability through testing in sample SME environments.

Click to know more

Security Topic 4:

Host-Based Intrusion Detection Systems (HIDS): Implementing and Evaluating Open-Source Security Monitoring in a Local Virtualized Environment

Research Aim

The main aim of this research is to assess/evaluate the effectiveness of Host-Based Intrusion Detection Systems (HIDS) in the detection of cyber attacks in a local virtual environment.

Research Objective

To conduct a thorough literature review on HIDS, focusing on their benefits and limitations.
To set up a virtualized environment using software like VirtualBox or VMware with various operating systems.
To install and configure multiple open-source HIDS solutions, such as Wazuh and OSSEC, on different virtual machines.
To simulate real-world cyber-attacks to test the detection capabilities of each HIDS tool
To analyze the performance and resource usage of each HIDS solution in both Windows and Linux environments.

Research Question

RQ1: How effective are Host-Based Intrusion Detection Systems in detecting security threats in a virtualized local environment?

Short Description of the idea

The project will explore the realm of Host-Based Intrusion Detection Systems, which monitor each host, with the focus of detecting unauthorized access or malware. Open-source host-based intrusion detection systems will be implemented such as Wazuh, OSSEC, and Tripwire, within a virtualized environment on either VirtualBox or VMware. By simulating real-world cyber-attacks on each tool, the project will evaluate the detection capability, quality of alerts generated, and resource usage of each system. The project aims to elucidate which tools best support local security monitoring and to offer practical guidance on their deployment. There will also be an examination of the disparity in performance of Windows and Linux OSes with these tools.

Research Methodology

Quantitative research methodology will be taken into consideration to successfully conduct this study. Under this methodology, experimental analysis will be conducted. But before conducting experimental analysis for the study, a detailed literature review will be conducted to identify gaps. In order to conduct this literature review difference databases including ScienceDirect, MDPI, and Google Scholar will be used. This review will cover the fundamentals of HIDS, their advantages, limitations, and deployment strategies. After completing the literature review, experimental analysis will be performed in a controlled virtual environment. The following steps outline this process

Set Up Virtual Environment
Install virtualization software (VirtualBox or VMware) on a local machine.
Create Virtual Machines
Set up multiple VMs with different operating systems (Windows 10/11 and various Linux distributions).
Install HIDS Solutions
Deploy multiple open-source HIDS tools (Wazuh, OSSEC, Tripwire) on the created VMs.
Simulate Cyber-Attacks
Conduct real-world attack simulations using tools like Metasploit and Hydra to test each HIDS's detection capabilities
Collect Data
Gather logs and alerts generated by each HIDS during the simulations
Analyze Results
Evaluate detection accuracy, response times, false positives/negatives, and resource usage across different operating systems

Click to know more

Security Topic 5:

Implementing and Evaluating Open-Source Security Information and Event Management (SIEM) in a Local Virtualized Environment

Research Aim

The main goal of this project is to implement and evaluate an open-source Security Information and Event Management (SIEM) solution in a local virtualized environment while assessing how well these tools function in detecting security incidents and processing logs in real-time.

Research Objective

To review the basic concepts of SIEM, including log aggregation and incident response.
To configure a virtual environment with virtualisation software such as VirtualBox or VMware.
To set up and install open-source SIEM solutions (Wazuh and Graylog).
To integrate various log sources, including firewalls and servers, for comprehensive monitoring.
To simulate security incidents and evaluate the SIEM's ability to detect them accurately.

Research Question

RQ1: How effective are open-source SIEM solutions in detecting security incidents in a local virtualized environment?

RQ2: What challenges arise when deploying SIEM in an on-premise setup, and how can these challenges be addressed?

Short Description of the Idea

The research aims to implement and evaluate open-source Security Information and Event Management (SIEM) solutions in a local virtualized environment. In the study, the main aim is to understand how well these tools can identify security incidents on information coming directly from the logs, firewalls, servers, and endpoints. With the development of a virtual lab using software such as VirtualBox or VMware, different operating systems will be set up, including Windows and Linux machines. Open-source SIEM tools that will be deployed for analysis and aggregation in real-time, include Wazuh, Graylog, and Elastic Stack. The project will also examine various cyber threats, such as brute-force attacks and malware execution, to test how well SIEM solutions can detect these incidents.

Research Methodology

To conduct this study, a Quantitative research methods will be used. A literature review will first be conducted using several academic databases including MDPI, research gate, IEEE Xplore, Google Scholar and ResearchGate. The review will include existing studies on open-source SIEM solutions and their effectiveness at detecting security incidents, as well as identifying existing gaps in the current knowledge.

Experimental analysis will be conducted after the literature review is completed in a virtual controlled environment. This process will involve creating a lab where virtualization tools like VirtualBox or VMware will be used to run several virtual machines representing different operating systems. The main focus will be on installing open-source SIEM tools like Wazuh or Graylog.

Upon the SIEM system setup, various log sources will be integrated into the system, and the sources will include firewall logs, Windows Event Logs, and Linux Syslogs. This configuration will be followed by execution of simulated cyber-attacks to assess the detection capabilities of the deployed SIEM solutions. Metrics collected at this stage will include alert accuracy, false positives, and resource utilization.

The data analysis will compare the performance of different SIEM tools in terms of their detection of the simulated incidents accurately. Results of this experimental analysis would assist the small enterprises with insightful evidence with regards to best practices in deploying SIEM solutions while addressing challenges that can be faced while implementation.

Click to know more

Security Topic 6:

Cloud-Based Network Intrusion Detection System (NIDS) for Threat Monitoring and Prevention

Research Aim

The aim of this project is to deploy a Network Intrusion Detection System (NIDS) within a cloud environment to monitor normal or malicious network traffic, and mitigate cyber threats in real-time. The system will leverage open-source tools to examine packet flows, recognize attack patterns, and notify administrators of possible risks. It will combine signature-based and anomaly-based detection approaches to increase the accuracy of threat detection.

Research Objective

To set up an AWS cloud infrastructure for the deployment of the NIDS.
To set up and configure an open-source NIDS tool (Snort or Suricata) in the cloud.
To monitor network traffic in real-time using the configured NIDS.
To implement a logging and alerting system to identify unauthorized activities.
To test the NIDS with various cyber-attack scenarios to evaluate its effectiveness.

Research Question

RQ1: How effective is a cloud-based NIDS at detecting real-time cyber threats?

RQ2: What challenges are faced when deploying NIDS in cloud environments?

Short Description of the Idea

The project aims to develop a Network Intrusion Detection System (NIDS) for a cloud environment (Amazon Web Services (AWS)). The objective is thereby continuous monitoring of the flow of network traffic that tends towards potential cyber threats as they occur. Such a system may include contemporary tools developed to monitor cyber traffic applied in intrusion detection like Snort or Suricata. By integrating these tools into a cloud environment, the project aims to enhance cybersecurity measures in dynamic and scalable dynamic environment settings. The system would include the monitoring of activities and alerting when suspicious activities occur, thus enabling fast responses from administrators in case of any incident. This alert will help the administrators to respond quickly to the incidents. Furthermore, the research will include NIDS performance and accuracy evaluation by implementing the NIDS against different cyber-attacks like DDoS attacks and unauthorized access attempts by the cybercriminals.

Research Methodology

In order to conduct this study, Quantitative research methodology will be followed. Under this Quantitative research methodology, experimental analysis will be conducted to assess the performance of the cloud-based NIDS. However, before conducting an experimental analysis, a literature review will be conducted to identify existing research gaps. This review will help in understanding current challenges and solutions related to deploying NIDS on platforms like AWS. Further, an experimental study using the implemented Network Intrusion Detection System (NIDS) will be followed after the literature review phase. The experimental phase begins by setting up an AWS environment to support NIDS tool operations on virtual machines (VMs). AWS Traffic Mirroring will serve as the basis for tool installation and configuration of Snort or Suricata to monitor network traffic. During the experimental phase, various cyber-attack scenarios will be simulated using Kali Linux and Metasploit tools. The evaluation process will determine how well the developed NIDS system identifies various threats. The data collected from the simulation will be analyzed to determine the system’s ability to detect threats as well as assess its rate of generating false alerts. For log management, the ELK stack (ElasticSearch, Logstash, Kibana) will be used to create visible log reports and SNS (AWS Simple Notification Service) will generate alerts.

Click to know more

Security Topic 7:

A Comparative Security Analysis of WPA2 and WPA3: Enhancements, Vulnerabilities, and Practical Implications

Research Aim

The research aims to conduct a comprehensive comparative analysis of the security enhancements in WPA3 over WPA2, examining past vulnerabilities in WPA2 and their relevance today, while performing practical penetration tests to evaluate the resilience of both protocols against common attacks.

Research Objective

To conduct a thorough review of literature regarding WPA2 and WPA3 security features.
To identify key vulnerabilities in WPA2 and evaluate if they persist in WPA3.
To examine the effects of WPA3 security improvements which include Simultaneous Authentication of Equals (SAE) and Protected Management Frames (PMF) and Enhanced Open networks.
To execute realistic attacks against WPA2 while also testing WPA3 vulnerabilities.
To execute realistic attacks against WPA2 while also testing WPA3 vulnerabilities.
To establishes practical guidelines and recommendations to enhance Wi-Fi security at homes and enterprises.

Research Question

RQ1: What security improvements does WPA3 introduce over WPA2, and how effective are they in mitigating known vulnerabilities?

RQ2: How does the implementation of Simultaneous Authentication of Equals (SAE) within WPA3 affects handshake security in a different way than WPA2-PSK?

Short Description of the Idea

The research evaluates security variances between Wi-Fi Protected Access (WPA) protocols particularly by studying how WPA3 delivers better protection compared to WPA2. Research will assess changes in real-world network security brought forth by WPA3 security enhancements which were developed to counteract brute-force attacks and de-authentication and open network exposure issues. The study performs a practical analysis to investigate WPA2 vulnerabilities through experimental tests in controlled settings. This research includes four stages which begin with literature review followed by experimental setup then penetration testing and analysis and finally results analysis with documentation. The experimental design comprises network setup with WPA2 and WPA3 security modes and hardware deployment of routers and Wi-Fi adapters alongside virtual or Linux platform simulation to conduct attacks. This phase of penetration testing consists of executing different types of attacks starting from handshake analysis through brute-force attack evaluation to de-authentication and management frame protection before finishing with probe request and evil twin attack remediation through the use of open-source tools Wireshark, Aircrack-ng, Hashcat and Bettercap. The assessment of WPA3 will compare its effectiveness against WPA2 weaknesses while creating reports that outline recommended best practices for building secure Wi-Fi networks.

Research Methodology

Quantitative experiments will be conducted to compare security features present in WPA2 and WPA3 protocols. A laboratory setup will be constructed to run WPA2 and WPA3 configurations over a wireless test network. Necessary hardware includes a dual-protocol router supported by a Wi-Fi adapter that has packet injection capabilities, and a Windows system together with either virtual machines or Linux systems such as Kali Linux or Parrot OS for attack simulation purposes. The experimental procedure will start with penetration testing and an analysis that will progress through several phases, which will include various attack simulation procedures. The collection of the Handshake data and comparison of the WPA2 and WPA3 protocols will be carried out by Wireshark along with Aircrack-ng to analyze how Simultaneous Authentication of Equals (SAE) in WPA3 has enhanced the authentication-security measures. The success rates of dictionary attacks against WPA2 and WPA3 will be evaluated through Hashcat and Aircrack-ng to evaluate their resistance to brute-force cracking. Using MDK3 and Aireplay-ng, de-authentication attacks against the two protocols will be conducted so that PMF in WPA3 may be assessed. Testing of probe request spoofing and Evil Twin attacks will assess WPA3's performance against vulnerabilities when compared to WPA2.

Machine Learning

Click to know more

Machine Learning Topic 1:

Automated Translation from Standard English to Simplified Technical English Using Large Language Models

Research Aim

The research aims at developing an automated system that translates complex technical documents from Standard English into Simplified Technical English to enhance accessibility for a non-expert audience.

Research Objective

To train a model that simplifies technical content while keeping its original meaning intact.
To assess the model’s effectiveness using readability measures and translation quality metrics.
To create an automated translation process that will aid accessibility in technical writing.

Research Question

RQ1: How can large language models be trained to effectively simplify technical documents into simplified technical English?

RQ2: What are the methods that could provide simplified language without loss of essential technical details?

Short Description of the Idea

In order to address challenges related to understanding complex technical documents, this study is centered on the development of an automated translation system providing translation from Standard English to Simplified Technical English. Traditionally, such translations rely upon human translators, which could make it a time-consuming and costly process, thus limiting accessibility on the part of non-experts. This project aims to automate the simplification process while preserving the original intent, using LLMs such as GPT-3 and T5. The innovative use of these advanced models permits better handling of technical jargons and formation of complex sentences, thereby bringing about better audience understanding. In addition, this study will use several measures for evaluating performance, including BLEU scores and readability, and checking simplification correctness to assess the quality and readability of generated content. It aims to bring technical documents closer to the public and, simultaneously, makes the field ready for an automated simplification tool for other forms of writing.

Research Methodology

Using the quantitative approach, the research design has been developed. Initially, the literature review was carried out to detect existing limitations. It was followed by an experimental approach applied to train models such as GPT-3 and T5 to translate complex technical texts into Simplified Technical English. A comprehensive dataset was collected from technical manuals and online forums to provide variations in the usage of technical language. Each model was tuned to learn about the context and jargon closely, focusing on pursuing a balance in the translation such that no information is lost in the process. The modeled outputs were evaluated based on BLEU scores for translation quality, Readability Index for readability, and Simplification Accuracy to ensure that essential components remain. Using this rigorous process, this project will produce a trustworthy and functional automated translation system to make technical writing accessible.

Click to know more

Machine Learning Topic 2:

Development of a Human Fall Detection System Using Ensemble Machine Learning Techniques

Research Aim

The primary goal of this research is to create a reliable and accurate human fall detection system with the integration of advanced ensemble machine learning in order to enhance safety for elderly subjects in a healthcare setting.

Research Objective

To combine multiple machine learning models to increase the accuracy of fall detection.
To thoroughly analyze sensor data from accelerometers and gyroscopes for effective fall identification.
To significantly decrease the occurrence of false positives and false negatives in the detection process.

Research Question

RQ1: To what extent can ensemble machine learning models be used to attain significant improvements in human fall detection?

RQ2: What combination of machine learning models, including decision trees, SVMs, and neural networks, yields best performance in detecting human falls?

Short Description of the Idea

This study proposes developing a human fall detection device based on an innovative ensemble of machine learning techniques to enhance the overall accuracy and reliability of systems that detect falls. Fall detection systems generally have one model at their basis, often leading to a false positive and false negative. In this project, a robust system is to be designed and developed whereby multiple algorithms such as decision trees, support vector machines, and neural networks are combined. Essentially, the approach will leverage ensemble techniques, such as random forests and gradient boosting, to extract the best of multiple models, allowing a comprehensive analysis of falling-related motion patterns. Key tools and techniques for the project include Python libraries such as scikit-learn for machine learning algorithms and TensorFlow/Keras for neural networks. In the end, this system aims to improve the safety of the elderly in healthcare settings by providing reliable and timely fall detection.

Research Methodology

The research methodology adopted in this study reflects a quantitative approach that allows systematic data collection and analysis. Firstly, to provide an overview of existing knowledge and to identify gaps in current fall detection technologies, a thorough literature review was conducted. Then, the experimental analysis employs publicly available datasets using accelerometers and gyroscopes data. This includes data pre-processing, feature extraction, and testing several ensemble machine learning algorithms in regard to fall detection. Numerous performance evaluation metrics including accuracy, precision, recall, and F1-score will be used to evaluate all models throughout the process. The methodology therein intends to develop an effective fall detection system after testing various combinations of models and tuning-in the parameters as it seeks to make the detection more reliable and reduce false positives and negatives to eventually identify and enable a quick response to falls under realistic situations.

Click to know more

Machine Learning Topic 3:

Development of a Data-Efficient Deep Learning Algorithm for Disease Classification in Chest X-Ray Images Under Data Scarcity Conditions

Research Aim

The primary aim of this research is to design more accurate and efficient deep learning models to classify diseases in chest X-ray images, especially where very limited labeled data is available.

Research Objective

To improve the accuracy of disease classification in chest X-ray using minimum labeled data.
To apply transfer learning and data augmentation techniques to maximize model performance.
To demonstrate the effectiveness of data-efficient deep learning models in the healthcare domain.

Research Question

RQ1: In what ways can transfer learning and data augmentation alleviate the challenges posed by limited data in chest X-ray classification?

RQ2: Which deep learning architectures yield the best performance for disease classification in environments with scarce data?

RQ3: How does the proposed model's classification accuracy compare to that of traditional models?

Short Description of the Idea

This research will be concentrated on creating an algorithm that is data-efficient based in deep learning, to enhance disease classification in chest X-rays, especially when a labeled dataset is limited. Conventionally, medical image classification has been dependent on extensive datasets in order to reach good performance, which can be difficult in the context of limited resources. This project will leverage advanced techniques such as transfer learning by utilizing pre-trained models like ResNet-50 and DenseNet to enhance performance despite data limitations. Furthermore, data augmentation techniques will be utilized to artificially increase the size of the dataset by introducing the following variations (i.e., rotation, flipping, and scaling). Combining these tactics aims to improve the robustness of the model while keeping classification accuracies high. The expected result will show that it is possible to obtain effective disease classification in medical images with even a lower number of labeled examples than traditional methods. This work will not only make a contribution to the field of medical diagnostics, but will also tackle major problems that are caused by data shortage in health informatics.

Research Methodology

The research methodology employed in this study is based on a quantitative approach.
The first step involves a comprehensive literature review to understand the current state of deep learning in medical imaging.
Experimental analysis is then conducted, which includes
- Data acquisition from publicly available databases e.g., NIH Chest X-ray 14.
- Data preprocessing and augmentation to enhance model robustness.
- Implementation of transfer learning using pre-trained models like ResNet-50 and DenseNet.
- Fine-tuning of the models on disease classification of chest X-ray images.
- Model performance evaluation via metrics, such as accuracy, AUC, and F1-score.
The analysis results are presented and compared with those from conventional models to show the superiority of the novel data-efficient deep learning algorithm.

Click to know more

Machine Learning Topic 4:

Assessing the Efficacy and Ethical Implications of Large Language Models in Emotion Detection: Practical Approaches to Addressing Limitations and Biases

Research Aim

The aim of this research is to provide an evaluation and understanding of the performance and ethical implications for large language models for emotion recognition from human language.

Research Objective

To perform a systematic analysis of the ethical implications of using LLMs for emotion detection, including the possible fallout from bias, misinterpretation, and impact on trust by users.
To rigorously evaluate the effectiveness and fairness of the LLMs in recognizing emotion, including finding out potential biases and disparities in their performance.
To propose strategies to mitigate bias in emotion detection.

Research Question

RQ1: What is the accuracy of the LLMs to detect emotion in varied text types including social media posts, emails and spoken conversations?

RQ2: What types of biases exist in LLMs for emotion detection and how do they reflect in model performance and decision making?

RQ3: How can the training and validation procedures of LLMs be modified such that their effectiveness in emotion detection could increase, and biases be decreased?

Short Description of the Idea

This analysis assesses the advantages and ethical ramifications of employing Large Language Models (LLMs) in the identification of emotions from human languages. This research seeks to gauge the performance of these models in the correct detection of emotions across different forms of text via a fine-tuning process of GPT-3 and BERT on emotion-labeled datasets. This discussion of ethics will highlight the critical opportunities to readdress inequities and disparities in LLMs that may result in misinterpretations and reduced trustworthiness. As the focus of this work is on the fairness and representativeness of such AI systems, it serves to direct the recommendations towards researchers and developers regarding remedies to the identified biases. Such a combination would guarantee more transparent AI models consistent with accountability and responsibility. Given its exploration of the dynamic with respect to LLMs, this study will contribute in genuine ways towards the improvement of the accuracy and ethics of AI systems.

Research Methodology

The type of research methodology employed in this study is exclusively quantitative, incorporated as an adjunct to a thorough literature review with experimental tests. A literature review will be conducted using different databases such as MDPI, Ieeee and research gate to find out relevant paper in order to identify the research gap before proceeding to experimental analysis where LLMs such as GPT-3 and BERT will be fine-tuned on emotion-labeled datasets such as ISEAR and EmotionX. In the experimental stage, the model performance, bias and ethical implications will be evaluated with the use of metrics, i.e., accuracy, F1-score and techniques for bias detections. This quantitative approach guarantees an objective assessment procedure with respect to the implementation of LLMs in emotion recognition, and at the same time offers relevant guidelines to reduce potential biases and increase fairness in such artificial intelligence systems. The methodological strength leads to trustworthy and generalizable results, thus contributing to the flow of acquired knowledge in the discipline.

Click to know more

Machine Learning Topic 5:

Development of a Predictive Visualization Tool for Understanding Housing Market Dynamics in England and Wales Using Data Science and Artificial Intelligence

Research Aim

The primary aim of this research is to develop a user-friendly predictive visualization tool, based on artificial intelligence and data science methodologies, to analyze and understand the housing market dynamics in England and Wales.

Research Objective

To leverage ML techniques to predict housing behavior trends in England and Wales.
To incorporate real-time data for up-to-date visualizations of the housing market.
To design an intuitive dashboard that enables stakeholders to conveniently analyze market trends.

Research Question

RQ1: How can machine-learning methods efficiently predict trends on the housing market based on the available data?

RQ2: What factors are most influential on changes in the housing market?

RQ3: How can real-time data be integrated into models to enhance the accuracy of housing market predictions?

Short Description of the Idea

This research project seeks to develop a predictive visualization tool to understand the housing markets in England and Wales, using artificial intelligence (AI) and data science approaches. By employing machine learning techniques such as ARIMA for time-series analysis and Random Forest for regression models, this project will forecast house prices and ascertain the key variables driving market behavior. The tool will integrate real-time data updates on housing trends, enabling users to see the most recent status of the housing market. A robust and easy-to-use interface will allow a variety of stakeholders, buyers, sellers, and investors, for instance, to make their decisions based on a reasonable level of knowledge. The available innovative tool will benefit the real estate sector by measuring model performance by such indicators as Mean Absolute Percentage Error (MAPE) and R². In the long run, the objective is for useable and meaningful real insights to help the user make sense of housing market shifts with increased clarity.

Research Methodology

Quantitative Methodology
The study will employ quantitative methodology for the analysis of housing market data.
Literature Review
A detailed literature review will be carried out to identify some of the studies and methodologies that have been put in place for housing market predictions.
Data Collection
Housing market data will be sourced from publicly available sources of information, including transaction records and economic indicators.
Experimental Analysis
Use the machine learning models ARIMA for time-series forecasting and Random Forest to assess the key market determinants.
Dashboard Development
Create a user-friendly visualization dashboard to display the findings and predictions in an accessible format.
Evaluation Techniques
Assess model performance using evaluation metrics such as Mean Absolute Percentage Error (MAPE) and R² to ensure the accuracy and reliability of predictions.

Click to know more

Machine Learning Topic 6:

Automated Code Style Checker for Enforcing Programming Assignment Guidelines in Student Projects

Research Aim

The aim of this research is to develop an automated tool that will check a student's coding exercise against the established programming standards to enhance the quality and consistency of coding.

Research Objective

To create a tool that identifies violations of coding style in a student's projects.
To promote student awareness of programming standards and best practices.
To automate the code review process in educational settings for efficiency.

Research Question

RQ1: How can static analysis effectively identify programming style violations in student code?

RQ2: How accurately is the automated system finding common errors in the students' submissions?

RQ3: What effect does this automated tool have on students' coding behaviors and compliance with guidelines?

Short Description of the Idea

The study addresses the issue of a common challenge within the educational environment regarding the consistency of coding practices in student assignments. It designs an automated code style checker to be implemented to ensure that the standards set by a course instructor are observed. This suggested project takes advantage of some of the static and natural language processing techniques in finding and reporting violations of coding style, from improper indentation to variable naming conventions or the use of out-of-date functions. By implementing BERT and spaCy, it aims to focus on effective syntactical and semantical understanding of the code so that the feedback provided is both factually accurate and also prompts student reflections toward better coding. The automated code checker offers students prompt and constructive feedback on their code, which will help them improve their code toward best practices and institutional standards. This new proposal is seen to free the instructors from tedious code reviewing while elevating student submissions to higher quality. Finally, it stimulates students' understanding of programming standards toward consistency and excellence.

Research Methodology

The methodology that is applied in this research work is of quantitative nature and has several steps à

Literature Review
A comprehensive view of current literature on coding standards, style guidelines, and the effectiveness of static code analysis tools in order to find out existing research gap.
Experimental Analysis
The development of the automated code checker will involve running tests against various coding style guidelines to identify possible violations.
Data Collection
This will consist of gathering materials from student submissions and publicly available code repositories to analyze the reality behind coding practices and violations.
Tool Development
The implementation process involves using high-level natural language processing libraries such as BERT and spaCy to analyze code style and syntactic issues.
Evaluation Metrics
The measurements applied to evaluate the working of this tool will be precision, recall, and F1-score, therefore measuring its ability to accurately detect coding violations.
Feedback Mechanism
It involve designing a system that gives students timely and actionable feedback on the violations detected to encourage improvement and adherence to coding standards.

Click to know more

Machine Learning Topic 7:

Design and Development of a Machine Learning-Driven Autonomous Inspection System for Detecting Defects in Offshore Wind Farm Turbines Using Image Analysis

Research Aim

The project aims to develop autonomous systems based on machine learning that can inspect offshore wind turbines for defects.

Research Objective

To design an autonomous inspection system for offshore wind turbines.
To utilize machine learning techniques for detecting structural defects in turbine components.
To evaluate the accuracy of the defect detection system using various performance metrics.
To analyze the challenges of using image analysis in offshore environments.
To improve the overall maintenance process of offshore wind turbines through automation.

Research Question

RQ1: How can machine learning techniques be effectively applied to detect defects in offshore wind turbine structures?

RQ2: What challenges are associated with using image analysis for defect detection in offshore environments?

Short Description of the Idea

The project proposes a novel solution for inspection of offshore wind turbines through a machine-learning-driven autonomous system. With advanced machine learning techniques like Convolutional Neural Networks and Transfer Learning methods, the proposed system will automatically identify anomalies such as cracks and corrosion without requiring human supervision. Data for training the model will be taken from open databases that include images of wind turbine inspections. Its performance will be assessed via selected measures, including accuracy, precision, and recall, to understand how well the model detects various defects.

Research Methodology

The quantitative research methodology will be used for the project. First, a literature review will be conducted with the goal of identifying the existing research gaps relevant to the field of automated inspection of offshore wind turbines. In order to conduct this literature review, different keywords including “Autonomous Inspection System”, “Machine Learning-Driven Autonomous Inspection System”, “Defect detection” will be used to find relevant papers. Additionally, it will provide insights into the current technologies and methodologies used for defect detection. This is followed by experimental analysis. The system will review pictures from drones or cameras to detect defects on turbine structures. CNNs and Transfer Learning will be used to fine-tune pre-trained models for successful defect detection. This will include preprocessing, which covers augmentation and normalization of a dataset, to provide the best performance for the mode. When the model is trained, it will be validated using quantitative metrics-including the model evaluation of accuracy, precision, and recall.

Click to know more

Machine Learning Topic 8:

Designing a Smart Image-Based Autonomous System for Automated Detection of Surface Defects in Metal Components Using Machine Learning.

Research Aim

The project aims to develop autonomous systems based on machine learning that can inspect offshore wind turbines for defects.

Research Objective

To develop a machine learning model that can automatically detect defects in industrial products.
To integrate computer vision techniques with machine learning for real-time surface defect detection in metal components.
To evaluate the effectiveness of the system using precision, recall, and F1-score metrics.
To optimize the system for use in dynamic industrial environments.
To provide insights on key image features that indicate defects in products.

Research Question

RQ1: How can machine learning models be used to effectively detect surface defects such as cracks, scratches, and rust on metal components in industrial images?

RQ2: What specific image features are crucial for identifying defects like cracks or scratches?

Short Description of the Idea

This project proposes a smart image-based system designed to detect defects in industrial products using machine learning techniques. In the system, defects such as cracks and scratches in metal components will be detected by images acquired by cameras. Through the convergence of computer vision and machine learning, the system will become fully autonomous and real-time quality analysis of the product can be achieved. The project will leverage existing industrial defect datasets like TID2008 and AID, which contain labeled images for training the model. Techniques such as image preprocessing and data augmentation will be employed to enhance the model's performance and robustness. The effectiveness of the system will be evaluated using metrics like precision, recall, and F1-score. The novel strategy is to minimize subjective inspection work in industrial environments and to achieve an increased quality of the products. Achieving real-time detection system optimization allows it to operate in diverse industrial settings.

Research Methodology

The project will be based on a quantitative approach. A literature review will be performed involving several academic databases (such as IEEE and Google Scholar) to discover the gap of the existing studies of defect detection in industrial products or metal components. This review will help establish a foundation for the experimental work by highlighting existing techniques and their limitations. After the literature review, experimental analysis will be carried out using advanced machine learning models like Convolutional Neural Networks (CNNs) or YOLO (You Only Look Once). The chosen models will be trained on preprocessed images from datasets, like TID2008 or AID. Image preprocessing techniques will include normalization and data augmentation to improve model accuracy and robustness. After training these models will be able to work autonomously and process incoming images from work settings in order to identify defects in real time. The performance of these models will subsequently be assessed on the basis of precision, recall and F1-score metrics in order to guarantee that they conform to the required quality for defect detection accuracy.

Click to know more

Machine Learning Topic 9:

Exploring the Efficacy of Bi-Directional Long Short-Term Memory (Bi-LSTM) Networks and Hybrid Approaches in Predicting Weather Forecasts and Detecting Climate Change Patterns

Research Aim

This project aims to investigate the effectiveness of Bi-Directional Long Short-Term Memory (Bi-LSTM) networks, in combination with hybrid approaches such as Convolutional Neural Networks (CNNs), in improving the accuracy of weather forecasts and detecting climate change patterns using historical weather data.

Research Objective

To apply Bi-LSTM networks for accurate weather predictions based on historical weather data.
To combine Bi-LSTM networks with CNNs to enhance prediction accuracy in detecting long-term climate change patterns.
To evaluate the model’s performance using common metrics such as Mean Absolute Error (MAE) and Root Mean Square Error (RMSE).
To conduct a comparative analysis with existing machine learning models to assess the benefits of the hybrid approach.

Research Question

RQ1: How does the combination of Bi-LSTM networks and CNNs improve the accuracy of weather predictions and climate change detection?

Short Description of the Idea

This project focuses on the need for accurate forecasting of weather and climate change pattern detection, two broad areas affecting societies globally. Existing methods often struggle to address the complex non-linear relationships in atmospheric data. In this regard, the project proposes to use Bi-Directional Long Short-Term Memory (Bi-LSTM) networks, which are particularly suited for learning temporal dependencies in time-series data. In this project, CNNs are also intended to be included in order to improve the data fed into the Bi-LSTM model. Historical weather data will be used to train the model on short-term forecasts and long-term climatic trends. Such sophisticated computation techniques are expected to significantly improve the predictive accuracy.

Research Methodology

In order to conduct the literature review, the research will following quantitative based experimental analysis. The experimental analysis phases consists of several phases, which are as follows. Firstly, the model training step will involve utilizing historical weather data where Bi-LSTM networks will extract temporal relationships within weather patterns. This research uses a dual method by allowing the CNN to extract features from the data while the Bi-LSTM network performs the time-series prediction task. The model will be trained on the dataset which include diverse weather features like precipitation, humidity, and temperature. Lastly, the trained model will be evaluated via MAE and RMSEperformance metrices to determine its accuracy when predicting both short-term weather conditions and long-term climate change predictions.

Click to know more

Machine Learning Topic 10:

Machine Learning Approaches for the Recognition and Classification of Urban Soundscapes

Research Aim

The aim of the project is to develop a machine learning model that can recognize and classify various sounds existing within urban settings as a way of enhancing the ability to monitor and manage urban noise effectively.

Research Objective

To design an automated system for analyzing urban sound recordings.
To categorize sounds into distinct classifications for real-time noise monitoring.
To improve understanding of urban noise patterns and their impacts.

Research Question

RQ1:What critical features in urban sound data enhance classification accuracy?

RQ2:What type of machine learning model performs the best in classifying urban sounds?

RQ3:How effectively do machine learning models generalize across different urban environments?

Short Description of the Idea

The project aims to develop advanced machine learning models that are able to accurately classify various sounds present in an urban environment, such as traffic noise, construction sounds, and human activity. Traditionally, urban noise monitoring relied heavily on manual observation and subjective analysis, which is often prone to biases and resource-intensive (Si et al., 2022). This project proposes an automated framework integrating the latest in machine learning technologies to streamline classification, hence offering a quicker and more objective means of dealing with urban noise. Using the UrbanSound8K dataset, which is a comprehensive collection with different urban sounds, the research will use audio feature extraction tools such as Librosa to convert the sound recordings into analyzable data. This new approach is intended to contribute more to sound classification and give important insights into urban sound dynamics for improved city planning and a better quality of life for urban centers.

Research Methodology

This research employs quantitative methodology for a thorough classifying of urban sounds. The initial phase entails an extensive literature review, which investigates the earlier undertaking in investigating urban sound recognition. This will allow the review of gaps and opportunities to further explore. Once concluded with the literature review, the research will move to experimental analysis, beginning with the collection of data from the UrbanSound8K dataset, which curates an assorted selection of urban sound recordings. A combination of preprocessing and feature extraction using the Librosa library will help transform audio signals into usable feature sets (Restack, 2025). At this point, two machine-learning models, SVMs and CNNs, will be fit for sound classification (Su et al., 2019). Performance evaluation via accuracy, precision, recall, and F-measure metric will help measure the efficacy of both models.

Reference

Si, Y., Xu, L., Peng, X. and Liu, A. (2022). Comparative Diagnosis of the Urban Noise Problem from Infrastructural and Social Sensing Approaches: A Case Study in Ningbo, China. International Journal of Environmental Research and Public Health, [online] 19(5), p.2809. doi:https://doi.org/10.3390/ijerph19052809.
Restack (2025). Audio Processing Techniques with Librosa | Restackio. [online] Restack.io. Available at: https://www.restack.io/p/commercial-ai-audio-apis-answer-audio-processing-techniques-librosa-cat-ai [Accessed 7 Feb. 2025].
Su, Y., Zhang, K., Wang, J. and Madani, K. (2019). Environment Sound Classification Using a Two-Stream CNN Based on Decision-Level Fusion. Sensors, [online] 19(7), p.1733. doi:https://doi.org/10.3390/s19071733.

Click to know more

Machine Learning Topic 11:

Voice-Based Detection of Depression Using Machine Learning Techniques

Research Aim

This study aims to investigate the application of a non-intrusive voice analysis-based system to detect the early signs of depression by analyzing voice features such as pitch, tone, and speaking patterns with the aid of machine learning methods.

Research Objective

To gather and preprocess voice recordings from established depression-related datasets.
To identify and extract important voice features which correspond to depressive states.
To implement machine learning models-SVM, LSTM-and evaluate performance in detecting depression using voice data.
To investigate comparison between voice analysis and standard practice for detection of depression.

Research Question

RQ1:How can machine learning approaches make it easier to recognize depressive patterns within speech?

Short Description of the Idea

This research examines the creative approach of using voice analysis to detect initial indicators of depressive symptoms. The traditional diagnosis of depression depends on both patient self-assessments and healthcare professional manual evaluations through semi-structured interviews while retaining potential limitations from subjective healthcare provider judgments along with masked emotional indicators (Mao, Wu and Chen, 2023). This project develops an intrusive system based on advanced machine learning methods to analyze vocal characteristics which include pitch tone speech rate and vocal intensity variation for depression detection. This project will leverage existing datasets like DAIC-WOZ and TESS, which contain audio samples representing different levels of depression (Anand and Tank, 2019). Key voice features will be extracted and processed, then fed into machine learning models such as Support Vector Machines (SVM) and Long Short-Term Memory (LSTM) networks (Zhang et al., 2016). Performance of the models will be evaluated against various metrics including accuracy, precision, recall and F1 score, ensuring robust performance through cross-validation. By focusing on vocal characteristics, this project aims to enhance traditional methods of depression detection, offering a novel tool for early intervention in mental health care.

Research Methodology

A quantitative research design will be used for this study to conduct a structured evaluation of voice-based detection methods for depression. To begin this investigation researchers will perform an extensive review of current research about voice analysis and mental health to establish research gaps. Subsequently, the research will involve experimental testing of well-known speech datasets., specifically, DAIC-WOZ and TESS before extracting vital features. The analysis of the acquired voice data will utilize both Support Vector Machines (SVM) together with Long Short-Term Memory (LSTM) networks. The evaluation of depressive symptom detection accuracy by these models will rely on following metrics, accuracy, precision, recall and F1-score.

Reference

Mao, K., Wu, Y. and Chen, J. (2023). A systematic review on automated clinical depression diagnosis. npj Mental Health Research, [online] 2(1), pp.1–17. doi:https://doi.org/10.1038/s44184-023-00040-z.
Anand, A. and Tank, C. (2019). Depression Detection and Analysis using Large Language Models on Textual and Audio-Visual Modalities. [online] Arxiv.org. Available at: https://arxiv.org/html/2407.06125v1.
Zhang, S.-X., Zhao, R., Liu, C., Li, J. and Gong, Y. (2016). Recurrent support vector machines for speech recognition. [online] pp.5885–5889. doi:https://doi.org/10.1109/icassp.2016.7472806.

Click to know more

Machine Learning Topic 11:

Voice-Based Detection of Depression Using Machine Learning Techniques

Research Aim

Research Objective

To gather and preprocess voice recordings from established depression-related datasets.
To identify and extract important voice features which correspond to depressive states.
To implement machine learning models-SVM, LSTM-and evaluate performance in detecting depression using voice data.
To investigate comparison between voice analysis and standard practice for detection of depression.

Research Question

RQ1:How can machine learning approaches make it easier to recognize depressive patterns within speech?

Short Description of the Idea

Research Methodology

Reference

Mao, K., Wu, Y. and Chen, J. (2023). A systematic review on automated clinical depression diagnosis. npj Mental Health Research, [online] 2(1), pp.1–17. doi:https://doi.org/10.1038/s44184-023-00040-z.
Anand, A. and Tank, C. (2019). Depression Detection and Analysis using Large Language Models on Textual and Audio-Visual Modalities. [online] Arxiv.org. Available at: https://arxiv.org/html/2407.06125v1.
Zhang, S.-X., Zhao, R., Liu, C., Li, J. and Gong, Y. (2016). Recurrent support vector machines for speech recognition. [online] pp.5885–5889. doi:https://doi.org/10.1109/icassp.2016.7472806.

Click to know more

Machine Learning Topic 12:

Automated Pothole Detection Using Image Processing and Machine Learning Techniques

Research Aim

The goal of this project focuses on designing an automated pothole detection system through a combination of state-of-the-art image processing and machine learning algorithms to improve overall road safety while decreasing maintenance costs.

Research Objective

To develop a highly effective machine learning model for evaluating road images to identify and assess potholes efficiently, facilitating quick maintenance procedures.
To analyze the effectiveness of various image processing techniques in detecting potholes across diverse road images.
To compare the accuracy and reliability of different machine learning models in identifying and classifying potholes, as well as assessing their severity in real-time.

Research Question

RQ1:How effective are image processing techniques in detecting potholes across various road images?

RQ2:Which machine learning models (CNNs or YOLO) provide the most accurate pothole detection and classification?

RQ3:Can the system reliably assess the severity of potholes in real-time?

Short Description of the Idea

This project proposes a state-of-the-art system for automated pothole detection using advanced image processing and machine learning techniques. Traditionally, road specialists relied on labor-intensive inspections to identify pothole degradation, making the process subjective and prone to errors (Tedeschi and Benedetto, 2017). This research aims to revolutionize pothole detection by employing automation to analyze road images in real-time for identifying and classifying potholes. Utilizing datasets such as the Pothole-600, the system will preprocess images using OpenCV to enhance visibility and apply machine learning models such as Convolutional Neural Networks (CNNs) and YOLO (You Only Look Once) for accurate detection (Bhavana et al., 2024). Performance will be measured based on evaluation metrics, such as accuracy, precision, recall, and F1-score. This innovative approach not only enhances road safety but also mitigates potential infrastructure damage.

Research Methodology

The methodology of the proposed study will be based on a quantitative design that aims at analyzing the effectiveness of automated pothole detection systems comprehensively. Firstly, systematic literature review will be conducted, studying existing work related to pothole detection in order to identify gaps in the research and providing a theoretical foundation. This will be closely followed by an experimental analysis to formulate a prototype detection model. The dataset will comprise real-world road images showing various pothole conditions. Images will be pre-processed through specialized computer-vision libraries, using OpenCV primarily. The resulting data will then be tested using different machine learning techniques, like convolutional neural networks (CNN) and YOLO (you only look once) (Cheng, 2020). Model performance will be evaluated using several metrics, including accuracy, precision, recall, and F1 score. This approach seeks to yield a robust understanding of the way in which cutting-edge technology might profoundly revolutionize the monitoring and maintenance of road infrastructures by serving as a methodological framework and a practical guide.

Reference

Tedeschi, A. and Benedetto, F. (2017). A real-time automatic pavement crack and pothole recognition system for mobile Android-based devices. Advanced Engineering Informatics, 32, pp.11–25. doi:https://doi.org/10.1016/j.aei.2016.12.004.
Bhavana, N., Kodabagi, M.M., Kumar, B.M., Ajay, P., Muthukumaran, N. and Ahilan, A. (2024). POT-YOLO: Real-Time Road Potholes Detection Using Edge Segmentation-Based Yolo V8 Network. IEEE Sensors Journal, 24(15), pp.24802–24809. doi:https://doi.org/10.1109/jsen.2024.3399008.
Cheng, R. (2020). A survey: Comparison between Convolutional Neural Network and YOLO in image identification. Journal of Physics: Conference Series, 1453, p.012139. doi:https://doi.org/10.1088/1742-6596/1453/1/012139.

Click to know more

Machine Learning Topic 13:

The main aim of the project focuses on developing a system which uses image processing together with machine learning methods to detect various plant diseases through leaf image evaluation.

Research Aim

Research Objective

To create an effective model for disease classification and early detection purposes in order to enhance agricultural management practices.
To implement advanced image processing techniques that accurately highlight plant disease symptoms in leaf images, improving the model's ability to identify and assess plant health.
To evaluate the generalization capabilities of the model across different plant species and disease types, ensuring reliable performance in diverse agricultural settings.

Research Question

RQ1:What is the level of effectiveness with which image processing methods detect plant disease symptoms in leaf images?

RQ2:Which machine learning models are most effective for classifying plant diseases based on extracted features from images?

RQ3:To what extent does the model generalize across diverse plant species and various disease types?

Short Description of the Idea

The main aim of the project is to develop a unique system for the automated detection of plant diseases through leaf image analysis. Traditionally, plant disease identification has involved expert visual inspection, which is slow and often inaccurate due to potentially subjective interpretation (Isinkaye, Olusanya and Singh, 2024). This research aims to speed up the detection and diagnosis process by automating it. This project will involve real-time crop monitoring with state-of-the-art deep learning algorithms such as Convolutional Neural Networks (CNNs) and ResNet that will assist farmers in the detection and treatment of crop diseases quickly (Debnath and Saha, 2022). This approach not only enhances crop yield but also improves overall farm management. The project uses large datasets, such as the PlantVillage dataset, applying image processing techniques such as feature extraction and data augmentation. The models will be evaluated using metrics including Accuracy, Precision, Recall, and F1-score to ensure reliable performance. Ultimately, this effort aims to bridge the gap between traditional methods and modern technology, facilitating more efficient and evidence-based management of agricultural diseases.

Research Methodology

The research methodology will be quantitative to achieve objectively better and consistent results. A comprehensive literature review will be performed first to study existing research on plant disease detection methods and identify gaps. The findings from this literature review will help to design the next experimental analysis. The research computes its results from the established PlantVillage dataset that features a wide variety of plant images coupled with disease classification labels. After completing the literature review phase, the project will execute experimental model development that deploys both Convolutional Neural Networks (CNNs) and ResNet architecture models. Proficient image classification tasks require these selected models. A thorough evaluation of this modeling work will use key metrics to measure accuracy while also determining precision, recall and F1-score. The research design incorporates comprehensive methods that will generate meaningful insights about automated agricultural disease detection abilities to advance data-driven crop management techniques.

Reference

Isinkaye, F.O., Olusanya, M.O. and Singh, P.K. (2024). Deep learning and content-based filtering techniques for improving plant disease identification and treatment recommendations: A comprehensive review. Heliyon, [online] 10(9), p.e29583. doi:https://doi.org/10.1016/j.heliyon.2024.e29583.
Debnath, O. and Saha, H.N. (2022). An IoT-based intelligent farming using CNN for early disease detection in rice paddy. Microprocessors and Microsystems, 94, p.104631. doi:https://doi.org/10.1016/j.micpro.2022.104631.

Click to know more

Machine Learning Topic 14:

Recognition of Animal Facial Expressions for Assessing Wellbeing and Health Using Image Processing and Machine Learning Techniques

Research Aim

A goal of this research focuses on building an invasive-free system through image processing integration with machine learning methods to evaluate animal expressions which enables their health and wellness assessment.

Research Objective

To design and implement a machine learning model that analyzes animal facial expressions to provide insights into their health and emotional condition, aiding in veterinary care.
To evaluate the effectiveness of various machine learning techniques in classifying animal facial expressions to identify health-related emotions accurately.
To develop a real-time monitoring system that can adapt facial expression analysis to different animal species, facilitating timely interventions based on detected emotions or health issues.

Research Question

RQ1:In what ways can animal facial expressions reliably indicate their health and wellbeing?

RQ2:Which machine learning techniques prove most effective for classifying animal emotions and health conditions?

RQ3:How can the system be adapted for real-time monitoring across various animal species?

Short Description of the Idea

This study intends to develop an innovative system for screening animal health status and wellbeing indicators based on facial expression analysis. Assessments of animal health have been considered difficult to accomplish using conventional methods because of their invasive nature (Kiani, 2022). This project employs state-of-the-art image processing technologies and will be applied with machine learning techniques, including CNN and ResNet to establish refined scanning that avoids invasiveness while providing accurate evaluations of animals' emotional and physical states. In addition, the Aff-Wild2 and Fera datasets will focus on facial landmark detection and emotion recognition methods to extract the most influential facial features related to health-related states (Arabian et al., 2024). The models will be evaluated using metrics such as accuracy, precision, recall, and F1-score to ensure reliability. The application of such innovation offers potential revolution in animal health monitoring and thus better animal care.

Research Methodology

This study employs a quantitative research methodology to explore the feasibility and effectiveness of the proposed system. Initially, the literature review will explore studies on animal facial expression analysis and existing research gaps in current literature. This will be followed by experimental analytical work directed at the development and assessment of the proposed system.

The researchers employ Aff-Wild2 alongside Fera because these datasets contain extensive collections of animal facial expression pictures.
Develop and train Convolutional Neural Networks (CNNs) and ResNet models concerning the classification of facial expressions and health-related indicators.
The results obtained will basically cater to a broad evaluation of the system in the analysis through the calculation of accuracy, precision, recall, and F1-score metrics.

Such an organized, systematic approach will provide this research with a reliable and accurate monitoring device of animal health.

Reference

Kiani, A.K. (2022). Ethical Considerations regarding Animal Experimentation. Journal of Preventive Medicine and Hygiene, 63(2 Suppl 3), pp.E255–E266. doi:https://doi.org/10.15167/2421-4248/jpmh2022.63.2S3.2768.
Arabian, H., Tamer Abdulbaki Alshirbaji, J. Geoffrey Chase and Moeller, K. (2024). Emotion Recognition beyond Pixels: Leveraging Facial Point Landmark Meshes. Applied sciences, 14(8), pp.3358–3358. doi:https://doi.org/10.3390/app14083358.

Click to know more

Machine Learning Topic 15:

Enhancing Face Age Detection: Evaluating Transfer Semi-supervised Regression Models for Effective Facial Analysis in Big Data Contexts

Research Aim

The aim of this study is to enhance the facial age detection accuracy by using transfer learning and the semi-supervised regression approach. This approach will be used to overcome the limitations of large datasets often containing missing or noisy data.

Research Objective

To improve the accuracy of age estimation based on facial features through a semi-supervised regression technique.
To utilize pre-trained models and adapt them for specific tasks related to age prediction.

Research Question

RQ1:How do transfer learning-based models, specifically VGG-Face and ResNet, compare in terms of accuracy and efficiency when applied to facial age detection tasks?

RQ2:How do semi-supervised learning approaches enhance the performance of facial age prediction models, as evaluated through Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) metrics?

Short Description of the Idea

The aim of this study is to improve the capability for facial age detection by adopting cutting-edge methods for machine learning, that is, transfer learning and semi-supervised regression models. Traditionally, age estimation from facial images relied heavily on handcrafted features and simplistic models, which often fell short in accuracy, especially when dealing with large-scale datasets that were noisy or incomplete (ELKarazle, Raman and Then, 2022). Using powerful "pre-trained models" such as ResNet, this work is aimed to adapt these architectures to age prediction tasks and thus provide dramatically better performances. Also, the incorporation of semi-supervised learning will allow the model to make full use of both the labeled and unlabeled data, overcoming the difficulty of insufficient data (van Engelen and Hoos, 2019). The main instruments and methods used in this study will involve well-proven deep learning pipelines (e.g., TensorFlow, PyTorch), and evaluation measures (e.g., Mean Absolute Error (MAE) to characterize the accuracy and robustness of the model.

Research Methodology

The quantitative nature of the research methodology has been used in this study. To begin, a systematic literature review has been performed to gain knowledge about the current schemes and technologies in facial age detection. Based on this foundation, experimental analysis will be conducted, with the aims of developing an experimental approach to transfer learning and semi-supervised learning methods. The course of action will be to employ prominent data sets, including IMDB-WIKI, for training and testing of the models (Ethz, 2015). In this experimental phase, pre-trained models will be fine-tuned for the age prediction task, allowing performance improvements to be observed. The models will subsequently be evaluated through Mean Absolute Error (MAE) to measure their efficiency in the presence of noisy and incomplete data. The quantitative analysis phase will lead to conclusion drawings about facial analysis refinement methods for big data applications based on identified trends and patterns.

Reference

ELKarazle, K., Raman, V. and Then, P. (2022). Facial Age Estimation Using Machine Learning Techniques: An Overview. Big Data and Cognitive Computing, 6(4), p.128. doi:https://doi.org/10.3390/bdcc6040128.
Ethz (2015). IMDB-WIKI - 500k+ face images with age and gender labels. [online] Ethz.ch. Available at: https://data.vision.ee.ethz.ch/cvl/rrothe/imdb-wiki/.
van Engelen, J.E. and Hoos, H.H. (2019). A Survey on semi-supervised Learning. Machine Learning, [online] 109. doi:https://doi.org/10.1007/s10994-019-05855-6.

Click to know more

Machine Learning Topic 16:

Enhancing Healthcare Delivery: The Development and Implementation of AI-Powered Virtual Assistants Utilizing NLP and Machine Learning Techniques

Research Aim

The purpose of this project is to develop an AI-based virtual assistant to enhance the delivery of healthcare services by utilizing natural language processing (NLP) and machine learning techniques to provide accurate and personalized responses to patient inquiries.

Research Objective

To undertake a literature review that identify gaps in AI applications in healthcare facilities and systems.
To develop an AI-based virtual assistant system which successfully handles healthcare patient inquiries stands as a project goal.
To train the assistant using medical dialogue datasets specifically designed for patient-provider conversations.
To implement advanced NLP models that improve the accuracy of the assistant’s responses.
To evaluate the assistant based on measure such as accuracy alongside response time.

Research Question

RQ1: How can NLP and machine learning be effectively used to create a virtual assistant for healthcare?

RQ2:What challenges arise in developing a virtual assistant that accurately understands medical queries?

Short Description of the Idea

The study works toward creating an AI virtual assistant system to handle healthcare patient inquiries efficiently. The assistant will utilize advanced natural language processing techniques and machine learning to comprehend and respond to medical questions (Yong and Zhang, 2023). Based on patient input, it will provide precise and personalized health recommendations, thus ensuring better access to healthcare for everyone. It is quite novel in its form and will facilitate human-like, context-aware interaction between a patient and the system, greatly enriching the patient experience. The assistant will train on medical dialogue datasets, such as the Medical Dialogue Dataset and MIMIC-III, allowing the assistant to learn from real examples pertaining to patient-provider interactions.

Research Methodology

Quantitative research methodology will be adopted for conducting this study (Ghanad, 2023). To begin with, a relevant literature review will be conducted through the online database ResearchGate, IEEE, and Google Scholar to determine existing research gaps in regard to AI and healthcare. An experimental analysis will follow the literature review. The AI-powered virtual assistants will be trained using medical dialogue datasets. This procedure includes advanced NLP methods based on BERT or GPT-3 models to interpret medical inquiries correctly. Upon training completion, the performance of the system will be assessed based on main parameters: accuracy of question-answering, response time, and user satisfaction.

Reference

Ghanad, A. (2023). An Overview of Quantitative Research Methods. [online] Doi.org. Available at: http://dx.doi.org/10.47191/ijmra/v6-i8-52.
Yong, H. and Zhang, L. (2023). Machine Learning and Natural Language Processing Algorithms in the Remote Mobile Medical Diagnosis System of Internet Hospitals. ACM Transactions on Asian and Low-Resource Language Information Processing. doi:https://doi.org/10.1145/3632172.

Click to know more

Machine Learning Topic 17:

Exploring the Predictive Power of Social Media Profiles on Voting Preferences Through Data Mining Techniques

Research Aim

The purpose of this research is to uncover social media profile characteristics that can identify voting preferences through data mining techniques. This will be accomplished through the collection and analysis of publicly available social media data, including text mining, sentiment analysis, social network analysis, and clustering, to correlate online behaviour with political leanings.

Research Objective

To collect publicly available social media data from Twitter and Facebook platforms.
To pre-process the collected data for analysis by cleaning and standardizing the formats.
To apply text mining techniques for extracting political themes and patterns from posts.
To conduct sentiment analysis to determine the political sentiment expressed in user interactions.
To analyse social networks to assess user influence on political behaviour and connections.
To implement clustering algorithms to group users with similar political preferences and behaviours.
To evaluate the effectiveness of analysis using metrics like sentiment score and cluster purity.

Research Question

RQ1:How can social media profiles predict voting preferences based on online interactions and what data mining techniques can uncover political views from social media activity?

Short Description of the Idea

Social media is a rich source of personal data that can shed light on voting behaviour (Poy and Schüller, 2020). The project will investigate ways in which data-mining techniques can identify patterns in social media profiles and predict voting preferences. The methodology involves collecting publicly available data from Twitter, Facebook, etc. and applying data-mining techniques to analyse patterns in relation to voters' political views and voting intentions. The project represents a new direction, as, unlike almost all previous work in this area, it attempts to make connections between behaviours in social media by analysing posts and connections, and voting preferences with the use of machine learning. Text mining, social network analysis, and clustering algorithms will be used to discover trends in individuals' political preferences. Sentiment analysis will help understand what political sentiment individuals expressed in their posts and comments, while network analysis will show how online connections may influence individuals during voting behaviour (Rita, António and Afonso, 2023). Clustering will pair users with similar political views based on their interactions and online behaviour. Performance metrics will be used to assess the significance of the solution.

Research Methodology

In order to conduct the research, a quantitative based experimental analysis will be followed along with the literature review approach to identify current gaps in the area. The experimental approach consists of diverse steps which are as follows. Firstly, the social media data will be retrieved from Twitter and Facebook through their public APIs or by utilizing dataset available on the Kaggle platform. The next step is to utilize text mining as well as sentiment analysis and social network analysis to study user posts as well as their social media interactions and network connections (Rueger, Dolfsma and Aalbers, 2023). The analysis will identify user groups using clustering algorithms to identify members with comparable political leanings through their social media actions. The method will use sentiment analysis to determine political sentiment within users' posts. Predictive model effectiveness will be evaluated through a combination of sentiment score evaluation and network centrality analysis and cluster purity measurement.

Reference

Poy, S. and Schüller, S. (2020). Internet and voting in the social media era: Evidence from a local broadband policy. Research Policy, 49(1), p.103861. doi:https://doi.org/10.1016/j.respol.2019.103861.
Rita, P., António, N. and Afonso, A.P. (2023). Social media discourse and voting decisions influence: sentiment analysis in tweets during an electoral period. Social Network Analysis and Mining, 13(1). doi:https://doi.org/10.1007/s13278-023-01048-1.
Rueger, J., Dolfsma, W. and Aalbers, R. (2023). Mining and analysing online social networks: Studying the dynamics of digital peer support. MethodsX, 10, p.102005. doi:https://doi.org/10.1016/j.mex.2023.102005.

Click to know more

Machine Learning Topic 18:

Social Media Sentiment Analysis: A Novel Approach to Brand Reputation Management

Research Aim

The aim of the research is to evaluate how social media sentiment analysis impacts brand reputation by analysing posts and reviews related to a brand.

Research Objective

To conduct literature on social media sentiment analysis and its effects on brand reputation management.
To apply the existing datasets like Sentiment140 and Twitter API for utility in sentiment analysis of social media data surrounding brand reputation.
To compare the efficiency of different NLP techniques, like VADER, TextBlob, and BERT classification for positive, negative, and neutral sentiments, on social media commentary.
To examine the proposed solution in terms of accuracy, precision, and F1 score and compare with traditional approaches.

Research Question

RQ1:How do different NLP models (such as VADER, TextBlob, and BERT) compare in effectiveness for classifying sentiment in social media posts?

RQ2:In what ways do shifts in sentiment over time correlate with specific marketing campaigns or public events, and how can these correlations inform strategies for improving brand reputation?

Short Description of the Idea

The rapid growth of social media interactions is making it challenging for businesses to monitor and manage their brand reputation in real-time (Rust et al., 2021). This research aims to investigate the intersection of sentiment analysis and brand reputation management using various natural language processing techniques on social media data. The research uses datasets such as Sentiment140 and models such as VADER, TextBlob and BERT to dive at their effectiveness in classifying the sentiment of social media posts. It will also aspect at how changes in sentiment relate to specific marketing campaigns and major public events, thus providing useful information for brand reputation management and strategic decision-making by businesses in today's fast-paced online environment.

Research Methodology

To carry out the research, the study will follow quantitative-based experimental analysis. During the analysis, posts, tweets, and reviews are expected to be collected from datasets such as Sentiment140 or Twitter API. With the help of named entity recognition techniques, better known by the acronym NLP, sentiment analysis will classify sentiments as positive, negative, or neutral using models such as VADER, TextBlob, or BERT (Jim et al., 2024). The system will analyse social media posts listed on brand standings over a period of time to see how sentiment changes and how those changes are linked to brand reputation. The multi-metric evaluation of performance in sentiment classification will include accuracy, precision, and F1 score. The analysis will extend to correlating sentiment trends with specific events, campaigns, or public perceptions that impact brand reputation.

Reference

Jim, J.R., Apon, M., Partha Malakar, Md Mohsin Kabir, Nur, K. and M.F. Mridha (2024). Recent advancements and challenges of NLP-based sentiment analysis: A state-of-the-art review. Natural Language Processing Journal, 6, pp.100059–100059. doi:https://doi.org/10.1016/j.nlp.2024.100059.
Rust, R.T., Rand, W., Huang, M.-H., Stephen, A.T., Brooks, G. and Chabuk, T. (2021). Real-Time Brand Reputation Tracking Using Social Media. Journal of Marketing, 85(4), p.002224292199517.

Cyber Security

Click to know more

Cyber Security Topic 1:

Penetration Testing in AWS Cloud Using Open-Source Tools: A Comprehensive Security Assessment Approach

Research Aim

The aim of this study is to carry out a penetration test in AWS, by using open-source tools. This will enable the identification of security weaknesses and misconfigurations in the cloud, and thus allows small and medium-sized enterprises (SMEs) to improve their security in the cloud.

Research Objective

To set up a testing environment in AWS with essential services like EC2, S3, and IAM.
To conduct an assessment to discover assets and identify any exposed resources.
To execute reconnaissance activities using open-source tools to map the AWS infrastructure and identify publicly accessible resources.
To scan AWS services for vulnerabilities and misconfigurations.
To simulate exploitation and privilege escalation to understand potential attack scenarios.
To analyze network traffic and logs for security events.
To provide actionable recommendations for improving AWS security.

Research Question

RQ1: What are the most common security misconfigurations found in AWS cloud environments?

RQ2: How effective are open-source tools in identifying vulnerabilities in AWS security?

Short Description of the Idea

This project focuses on strengthening the security of AWS via penetration testing with the use of open-source tools. As more companies move to the cloud, it becomes increasingly relevant to be aware of the security risks of these environments. The motivation of this project is to help small and medium-sized enterprises show that very effective security assessments can be performed without expensive commercial tools. The project will set up a controlled AWS environment with EC2, S3, IAM, and Lambda in place and a number of tests will be run to discover vulnerabilities. The test consists of reconnaissance to identify exposed resources, scanning to check for any misconfigurations, and real-world attack simulations to establish its risks. The findings will be documented in a report.

Research Methodology

After the literature review, the experimental analysis begins with the setting up of the AWS testing environment, using the Free Tier account. The essential services like EC2 instances, S3 buckets, IAM roles, and Security Groups will be deployed.

Reconnaissance will be performed next, based on tools like CloudMapper and ScoutSuite, for mapping the infrastructure and identifying publicly exposed assets. Then, the vulnerability scanning will come into play based on tools like Nmap and OpenVAS for detecting misconfigurations in AWS services.

When these vulnerabilities are spotted, exploitation techniques will be simulated using frameworks like Metasploit and Pacu. In this phase, the ease with which an attacker can gain unauthorized access or escalate privileges within the environment is assessed. Finally, network traffic analysis will be executed by using tools like Wireshark and Suricata to identify any malicious activities occurring.

Click to know more

Cyber Security Topic 2:

Deploying and Evaluating Cloud-Based Honeypots for Threat Intelligence and Attack Detection

Research Aim

The purpose of this project is to deploy and assess cloud-based honeypots to learn more about cyber threats. The data collected from these honeypots will be analyzed in order to strengthen security measures for cloud environments.

Research Objective

To review the fundamentals of honeypots and their significance in cloud security.
To select and set up open-source honeypots suitable for deployment in a cloud environment.
To set up honeypots on a cloud platform with appropriate network settings.
To simulate real-world cyber attacks to evaluate the effectiveness of the deployed honeypots.
To analyze the collected data to identify attack patterns and provide recommendations.

Research Question

RQ1: How effective are cloud-based honeypots in detecting various cyber threats?

RQ2: What common attack techniques are observed in cloud environments?

Short Description of the Idea

This project intends to deploy cloud-based honeypots to detect and analyze cyber threats targeting cloud infrastructures. Honeypots are decoy systems designed primarily to deceive an attacker and collect data on their attack methods and techniques. The target of the project is to simulate different attack surfaces like SSH, web applications, and databases using open-source honeypots like Cowrie and Dionaea. The data is expected to give insight on attacker behavior and identify some of the more common attack patterns. Further, a cloud environment will be setup up to enable the proper configuration of networks for maximum monitoring effectiveness. The honeypot threat detection capabilities will be evaluated through simulated attacks conducted in a cloud-based environment.

Research Methodology

The quantitative research methodology for this project follows a structured approach which combines literature review and experimental analysis. A literature review will be conducted using various academic databases, like ScienceDirect, MDPI, IEEE, and ResearchGate, to identify existing studies on honeypots and their effectiveness in cloud environments. This review will help to identify gaps in current knowledge that this project aims to address. After conducting the literature review, experimental analysis will take place. The implementation involves deploying honeypots through tools like Cowrie or Dionea in a cloud environment on platforms like AWS or Azure for attack simulation along with attacker behaviour and pattern analysis.

Achieve High Distinction grades in your MSc dissertation with expert guidance from our academic writers. Contact us today for tailored dissertation services in AI, cybersecurity, and networking research, ensuring a well-structured, impactful, and academically rigorous dissertation!

Guide to Detailed Project Proposal of University of Hertfordshire

Here is the guide for Detailed Project Proposal (DPP) of the University of Hertfordshire, which outlines the research topic, objectives, methodology, and expected outcomes. We serve as a roadmap for your dissertation, ensuring clarity and feasibility before full-scale research begins.

Research Title:

Craft a title that captures the essence of your project, clearly indicating its focus and scope. Aim for clarity and conciseness to engage your audience from the start.

Example for Machine Learning:

Automated Translation from Standard English to Simplified Technical English Using Large Language Models

Example for Networking:

Automated Firewall CE Self-Assessment Tool Using Wazuh and pfSense

Example for Cyber Security:

Penetration Testing in AWS Cloud Using Open-Source Tools: A Comprehensive Security Assessment Approach

Aim:

Write a brief statement that outlines the primary goal of your research. The aim of your research should clearly articulate the purpose of your project in a single statement, ideally beginning with an infinitive verb (e.g., “to determine, “to investigate”, “to develop”), followed by an action that captures the essence of your inquiry. This ensures that the aim addresses what you intend to investigate, how you plan to approach it, and remains concise while guiding your research direction.

Example for Machine Learning:

Example for Networking:

Example for Cyber Security:

Penetration Testing in AWS Cloud Using Open-Source Tools: A Comprehensive Security Assessment Approach

Objectives:

When crafting objectives for your experiment or research project, it’s important to articulate specific goals that detail what you intend to achieve. Each objective should begin with the word “to”, followed by an action verb that clearly defines the task at hand. Examples of effective action verbs include “to review”, “to analyze”, “to design”, “to compare”, “to measure”, and “to evaluate”. This structure helps in outlining the steps needed to fulfill your aim and provides clarity on the direction of your study.

Example for Machine Learning:

To train a model that simplifies technical content while keeping its original meaning intact.
To assess the model’s effectiveness using readability measures and translation quality metrics.
To create an automated translation process that will aid accessibility in technical writing.

Example for Networking:

To set up a virtual environment to simulate different network scenarios.
To integrate Wazuh and pfSense for collecting and analyzing firewall logs and configurations.
To develop automated scripts for assessing firewall configurations against CE standards.
To analyze collected data against CE standards using data analysis tools.
To generate comprehensive reports summarizing assessment results and providing actionable recommendations.

Example for Cybersecurity:

To set up a testing environment in AWS with essential services like EC2, S3, and IAM.
To conduct an assessment to discover assets and identify any exposed resources.
To execute reconnaissance activities using open-source tools to map the AWS infrastructure and identify publicly accessible resources.
To scan AWS services for vulnerabilities and misconfigurations.
To simulate exploitation and privilege escalation to understand potential attack scenarios.
To analyze network traffic and logs for security events.
To provide actionable recommendations for improving AWS security.

Research Questions:

To formulate effective research questions, first identify the specific issue or problem that your research will address. Each question should begin with interrogative words such as "what," "how," "why," or "to what extent." Ensure that the questions are clear and focused; they should be neither too broad nor too narrow, analytical rather than descriptive, and framed to require in-depth exploration rather than simple yes-or-no answers. Additionally, make sure that the questions align with your study’s purpose and include sufficient material for research while remaining within the scope of your study.

Example for Machine Learning:

RQ1. How can large language models be trained to effectively simplify technical RQ2. documents into simplified technical English?
RQ2.What are the methods that could provide simplified language without loss of essential technical details?

Example For Networking:

RQ1. How effective is the automated tool in identifying and assessing firewall configurations compared to manual assessments?

Example for Cybersecurity:

RQ1. What are the most common security misconfigurations found in AWS cloud environments?
RQ1.RQ2. How effective are open-source tools in identifying vulnerabilities in AWS security?

Short Description of Your Idea:

Provide a concise overview of your project concept, highlighting its significance and relevance. To create an effective short description of our project concept, start by clearly stating the main objectives in one to two sentences, identifying the issue it addresses. Briefly explain the existing problem or challenge with current methods and highlight its importance. Present your solution, detailing how it can resolve the identified problem and any unique aspects that set it apart from existing solutions. Articulate the benefits of your project, such as improved efficiency or accuracy, and conclude by discussing its broader relevance. Explain who will benefit from your work and why it is important in your field or industry. This section should give readers a clear picture of your research’s importance.

Example for Machine Learning:

Example for Networking:

Example for Cybersecurity:

Methodology:

Outline the approaches and techniques you plan to employ to conduct your research. Firstly, select an appropriate approach based on your research question. For research involving quantifiable data and patterns, use a quantitative methodology with surveys or experiments and statistical analysis. If exploring motivations or perceptions, adopt a qualitative methodology through a literature review for in-depth insights. For a comprehensive analysis, employ a mixed methodology that combines both quantitative and qualitative approaches. Clearly outline your data collection methods, analysis techniques, and any tools or frameworks to ensure a systematic and robust investigation.

Example for Machine Learning:

Example for Networking:

Example for Cybersecurity:

The research will adopt a quantitative approach. A literature review will be performed by examining various databases, such as ScienceDirect, Google Scholar, etc., firstly to ascertain the existing gaps in the literature regarding penetration testing in AWS environments. The review will provide a base for understanding the current challenges and effective practices. After the literature review, the experimental analysis begins with the setting up of the AWS testing environment, using the Free Tier account. The essential services like EC2 instances, S3 buckets, IAM roles, and Security Groups will be deployed. Reconnaissance will be performed next, based on tools like CloudMapper and ScoutSuite, for mapping the infrastructure and identifying publicly exposed assets. Then, the vulnerability scanning will come into play based on tools like Nmap and OpenVAS for detecting misconfigurations in AWS services. When these vulnerabilities are spotted, exploitation techniques will be simulated using frameworks like Metasploit and Pacu. In this phase, the ease with which an attacker can gain unauthorized access or escalate privileges within the environment is assessed. Finally, network traffic analysis will be executed by using tools like Wireshark and Suricata to identify any malicious activities occurring.

Reference:

Compile a list of sources that inform and support your project. This should include journals, articles, books, and any other relevant materials that demonstrate your research foundation. Ensure that you use a consistent reference format, such as Harvard, IEEE, or APA 7th edition. This will enhance the credibility of your work and allow readers to find the original sources easily. To streamline the citation process, consider using tools like Crowjack’s citation generation, which can automatically format your references according to your chosen style.

Detailed Project Proposal Topics for MSc 2025

Security

Research aim

Research Question

Research questions

Short Description of the idea

Research Methodology

Research aim

Research Objective

Research Questions

Short Description of the idea

Research Methodology

Research aim

Research Objective

Research Questions

Short Description of the idea

Research Methodology

Research Aim

Research Objective

Research Question

Short Description of the idea

Research Methodology

Research Aim

Research Objective

Research Question

Short Description of the Idea

Research Methodology

Research Aim

Research Objective

Research Question

Short Description of the Idea

Research Methodology

Research Aim

Research Objective

Research Question

Short Description of the Idea

Research Methodology

Machine Learning

Research Aim

Research Objective

Research Question

Short Description of the Idea

Research Methodology

Research Aim

Research Objective

Research Question

Short Description of the Idea

Research Methodology

Research Aim

Research Objective

Research Question

Short Description of the Idea

Research Methodology

Research Aim

Research Objective

Research Question

Short Description of the Idea

Research Methodology

Research Aim

Research Objective

Research Question

Short Description of the Idea

Research Methodology

Research Aim

Research Objective

Research Question

Short Description of the Idea

Research Methodology

Research Aim

Research Objective

Research Question

Short Description of the Idea

Research Methodology

Research Aim

Research Objective

Research Question

Short Description of the Idea

Research Methodology

Research Aim

Research Objective