Master of Science in Applied Computer Science

Permanent URI for this collectionhttps://192.168.0.29/handle/123456789/128

Browse

Now showing 1 - 6 of 6

Investigating the threat of adversarial machine learning attacks on AI-powered security information and event management systems
(Adventist University of Africa, 2025-05) Kikandi Safari Isaac
In the evolving landscape of cybersecurity, Security Information and Event Management systems have emerged as the nerve centers of modern defense, correlating logs, detecting anomalies, and delivering alerts leveraging AI-powered intelligence. However, as organizations rely more on these machine learning components, a new threat arises—AML. These attacks don’t crash systems or flood networks, they manipulate data just enough to fool the AI into silence or confusion. This subtlety makes them both dangerous and difficult to detect. This research set out to simulate such a scenario by incorporating two neural network architectures: a Feedforward Neural Network and a Convolutional Neural Network as detection engines of a SIEM setting. Using a cleaned and balanced sample of the CICIDS2017 dataset, the models were trained and then tested by two evasion attack frameworks: FGSM and Projected Gradient Descent. These attacks, selected for their computational efficiency and widespread use in adversarial research, were applied under white-box conditions using the Adversarial Robustness Toolbox. The results told a striking story: the FGSM attack reduced the FFNN’s detection rate from 95% to just 4.6%, while the PGD attack caused a substantial decline in the CNN’s recall, dropping it to 25%. While the CNN fared better under FGSM, it responded by flagging benign traffic as malicious, raising the risk of alert fatigue. Metrics such as precision, recall, and attack success rates clearly illustrated how easily an AI model’s performance could be compromised with subtle adversarial effort. These findings confirm that without proper defenses, AI-driven SIEMs are critically exposed and the study recommends embedding robust countermeasures such as adversarial training, anomaly detection, and model hardening to safeguard these systems. As cyber threats grow smarter, so too must our defenses be, not only in strength, but in foresight.
Towards a culturally aware user interface evaluation instrument for virtual learning platforms: A case of Moodle in Kenya
(Adventist University of Africa, 2025-05) Nyanamba, Brenda Nyangweso
The growing utilization of Virtual Learning Platforms (VLPs) in higher education, especially in Kenya, has underscored the necessity for user interfaces that are both technically functional and culturally attuned. Although current usability evaluation techniques prioritize technical and educational elements, they frequently neglect the cultural aspects that influence user experience, particularly in environments like Kenya. This work aimed to address that gap by creating and verifying a culturally sensitive user interface evaluation tool specifically designed for assessing VLPs, particularly Moodle, the most prevalent platform in Kenyan colleges. The research was directed by three objectives: (1) Identify cultural dimensions influencing Kenyan users’ UI preferences. (2) Adapt existing UI evaluation tools to incorporate these dimensions. (3) Test the instrument’s efficacy in evaluating Moodle’s usability in Kenya. A hybrid framework was created by combining recognized usability heuristics, pedagogical usability principles, and cultural elements based on Hofstede’s model and intercultural design approaches. The tool underwent evaluation by five specialists in usability, instructional design, and Kenyan culture. The feedback was assessed and incorporated to enhance the instrument, which was subsequently utilized in a pilot review of Moodle. The findings indicated that the interface expectations of Kenyan learners are significantly shaped by cultural characteristics, including high power distance, collectivism, moderate uncertainty avoidance, and achievement incentive. The test successfully revealed culturally unique usability concerns frequently neglected in traditional assessments, including the necessity for localized language, prominent teacher supervision, and community-oriented learning aspects. The findings indicated that including cultural factors into UI assessment yields enhanced understanding of learner involvement and system effectiveness. The study recommends the implementation of culturally relevant evaluation tools in the design and implementation of VLPs in Kenya. The proposed instrument functions as a valuable tool for e-learning designers, educators, and policy-makers aiming to improve cultural inclusion and user experience in virtual learning environments across many educational contexts.
Design and implementation of an online platform for integration and analyzing multivariate multisource malaria data
(Adventist University of Africa, 2023-04) Ochola, Micah Asuke
One of the common public health problems reported by the World Health Organization (WHO) in the African Region is malaria, where the burden of the disease is highest globally. The greatest challenge experienced in the fight against Malaria is, surveillance, which leads to early detection and treatment, and is crucial for reducing transmission and preventing deaths. Malaria surveillance includes gathering, analyzing, and interpreting malaria-related data. Though there exist many facilities with Malaria data, the collection and integration of data from different sources has been a major challenge that needs to be addressed. The proposed solution is aimed at the development of an online solution that can be used to collect malaria data from multiple sources including hospitals, drug stores and weather stations in various formats and aggregated into a format that can further be used in the prediction of malaria outbreak. From the results, the system collects data from hospitals and drug stores, which is then integrated with weather data. The generated data was used to train a machine learning model, as a proof of concept to validate that it can it be used to predict malaria outbreaks. This solution does not only solve the problem of data collection and integration but also ensures timely actions are taken in cases of out breaks. The implementation of this solution therefore significantly improves on the current practices by ensuring that hospital records and over the counter sale of drugs are reported electronically, daily and in real-time as opposed to manually and weekly. The solution also introduces the use of multi-source data in the analysis of malaria outbreaks rather than only focusing on hospital records as the only source of information for outbreak detection. Further to this, the project has the potential to contribute to the WHO Global Malaria Technical Strategy 2016-2030, as early detection and treatment of malaria are essential for reducing the burden of the disease. The methodology and system produced in this study can be used in other regions to improve malaria surveillance and outbreak prediction.
Combining Meta-Heuristic Technique and Neural Networks to Detect Internet of Things Network Attacks
(Adventist University of Africa, 2023-03) SAMBA, Steve
The Internet of Things (IoT) refers to ordinary objects equipped with wearable sensors and batteries that can communicate over the internet and perform predefined actions. These devices are evading our everyday lives in many ways. It is now possible to sense temperature, and heart rate with a smartphone, while cloud applications can monitor security systems or smart home equipment. Consequently, IoT networks have simplified life. However, the growing popularity of Internet of Things devices poses security concerns that need attention. For instance, attackers may target IoT networks for several reasons, including a quest for personal, medical, or financial information and espionage. In certain circumstances, these attacks can have severe repercussions for people's lives. Others may face extortion, damaged reputation, impersonation, fraud, or financial damages. Detection is critical for defending IoT networks and avoiding the negative repercussions of cyberattacks. Detection consists of identifying assaults before they occur. Numerous writers have examined the security of IoT networks and proposed promising solutions based on machine learning. This research investigated how efficient could combining Neural Networks and Metaheuristic technique be in detecting IoT network attacks. To address that concern this study proposed a novel method that integrates neural networks for attack classification and Particle Swarm Optimization, a metaheuristic strategy for feature selection and hyperparameter tuning. The outcomes of the suggested strategy using two different IoT data sets, namely the BaIoT and the CICIDS 2017 datasets yielded accuracy scores of 98% and 99.95% for multiclass classification. The binary categorization was nearly flawless. Furthermore, this study revealed the potential of CNN, MLP and FFNN when dealing with classification problems for IoT environments. The study also highlighted interesting future venues for improving IoT network security, such as deployment, training models with higher quality datasets, or even tweaking more parameters.
Combining Meta-Heuristic Technique and Neural Networks to Detect Internet of Things Network Attacks
(Adventist University of Africa, 2023-03) Samba, Steve
The Internet of Things (IoT) refers to ordinary objects equipped with wearable sensors and batteries that can communicate over the internet and perform predefined actions. These devices are evading our everyday lives in many ways. It is now possible to sense temperature, and heart rate with a smartphone, while cloud applications can monitor security systems or smart home equipment. Consequently, IoT networks have simplified life. However, the growing popularity of Internet of Things devices poses security concerns that need attention. For instance, attackers may target IoT networks for several reasons, including a quest for personal, medical, or financial information and espionage. In certain circumstances, these attacks can have severe repercussions for people's lives. Others may face extortion, damaged reputation, impersonation, fraud, or financial damages. Detection is critical for defending IoT networks and avoiding the negative repercussions of cyberattacks. Detection consists of identifying assaults before they occur. Numerous writers have examined the security of IoT networks and proposed promising solutions based on machine learning. This research investigated how efficient could combining Neural Networks and Metaheuristic technique be in detecting IoT network attacks. To address that concern this study proposed a novel method that integrates neural networks for attack classification and Particle Swarm Optimization, a metaheuristic strategy for feature selection and hyperparameter tuning. The outcomes of the suggested strategy using two different IoT data sets, namely the BaIoT and the CICIDS 2017 datasets yielded accuracy scores of 98% and 99.95% for multiclass classification. The binary categorization was nearly flawless. Furthermore, this study revealed the potential of CNN, MLP and FFNN when dealing with classification problems for IoT environments. The study also highlighted interesting future venues for improving IoT network security, such as deployment, training models with higher quality datasets, or even tweaking more parameters.
A compensatory approach to anti-virus shortfalls
(Adventist University of Africa, School of Postgraduate Studies, 2020-05) Ongaro, Tom Ongaga
Computer systems security has become an increasingly important field. In the quest to provide the much-needed security many options exist. Systems have however continued to suffer attacks from malware despite the existing controls that have been put in place. One such control is the use of Anti-viruses which are widely used in many systems. Today malware exists that can bypass anti-viruses and cause harm to systems. Many controls exist to try to combat malware infiltration. Organizations and small businesses may not always be in a position to choose the best option for their environment when it comes to dealing with malware. They may not also be able to configure system security tools that may be available to deal with malware detection and prevention. One freely available tool is Sysmon. Sysmon logs critical events in a windows environment and can send them out for further analysis and classification. This research seeks to understand why some malware can bypass anti-viruses and seeks to close the gap by providing tangible recommendations. The end goal provides results that can be adopted by anyone to try to identify malicious activity in their systems by using freely available tools.

Browse

Recent Submissions