Pusat Pengajian Sains Komputer - Tesis

Browse

Now showing 1 - 5 of 487

A Sliding Adaptive Beta Distribution Model For Concept Drift Detection In A Dynamic Environment
(2025-06)
Angbera, Ature
Machine learning models deployed in data streaming environments often suffer from concept drift, where the underlying data distribution changes over time, leading to performance degradation. Detecting and adapting to these shifts in real time is crucial to maintaining model accuracy and reliability. This study introduces the Sliding Adaptive Beta Distribution Model (SABeDM), a novel approach for concept drift detection and adaptation in dynamic data streams.
A Hybrid Multi-Tier Approach For Iot Botnet Detection And Enhanced Risk Assessment
(2025-01)
Ali, Mashaleh Ashraf Sulieman
The proliferation of internet of things (iot) devices has led to new cybersecurity challenges. A significant issue is the increasing occurrence of iot botnets, which refers to networks of compromised iot devices like routers, ip cameras, and smart appliances. These compromised entities are strategically utilized to carry out various cyber threats, including but not limited to distributed denial of service (ddos) attacks, data exfiltration, and network reconnaissance. Identifying iot botnets has unique issues due to the constrained resources of the devices involved. This research contributes significantly by identifying the active phase of the iot botnet attack life cycle and enabling flexible evaluation of attack severity levels through an ensemble model stacking and boosting via a soft voting system integrated with a fuzzy logic-based risk assessment methodology optimized by particle swarm optimization. This provides a basis for security teams to allocate resources efficiently, enabling a proactive and dynamic cybersecurity defense against iot botnet threats. A realistic and representative iot dataset was also generated, simulating the iot botnet lifecycle and incorporating the most recent attacks on iot ecosystems. The proposed approach significantly advances iot security by enabling precise detection of botnet activities and proactive threat mitigation. The integration of ensemble learning, fuzzy logic, and pso offers a dynamic solution that adapts to evolving cyber threats, ensuring targeted, efficient responses and safeguarding network integrity.
Hybrid Machine Translation Using Malay-English Language Parallel Text Extraction From Comparable Text
(2024-12)
Yeong, Yin Lai
Machine translation (mt) investigates the approaches to translate a text from a source language (sl) to a target language (tl). Parallel text is the resource that is essential for building the translation model of an mt system. A parallel text is a text and its translation in one or more languages. Nevertheless, there are not many parallel texts that are freely available. Thus, a few directions were explored and investigated in this thesis to improve the translation quality despite the limited parallel text. Firstly, we analysed using linguistic information in machine translation to compensate for the lack of data for training. Secondly, we studied the problem of acquiring a parallel text from comparable texts. Comparable texts are similar texts in different languages that may be independently produced. Thirdly, we investigated the architecture of statistical machine translation (smt) and neural machine translation (nmt) to combine the strength of both systems. This study was carried out using english-malay machine translation in the news domain and computer science domain. For the first problem, we propose to use affixation and part-of-speech information to build a translation model. We improve the bleu score from 13.40% to 15.41% using 315,194 parallel texts. In the second problem, we propose an algorithm to extract parallel sentences and parallel fragments/subsentences from comparable texts. The approach finds matching comparable texts. Then, a sentence aligner and a classifier are used to align the sentences in the comparable text.
A Framework For Behavioural Intentions To Adopt Cryptocurrency Among Public University Students In The Kingdom Of Saudi Arabia
(2024-09)
Saeed A., Alomari Ali
By examining individual intentions, this study seeks to provide valuable insights into the behavioural drivers and barriers affecting cryptocurrency adoption. The study used a quantitative research method with developing a survey questionnaire. Using a purposive sampling technique, the study targeted the students in the Saudi Arabian universities. The study used SPSS software for demographic and descriptive statistics and Smart PLS for testing validity, reliability and research hypotheses.
Graph-Based Algorithm With Self-Weighted And Adaptive Neighbours Learning For Multi-View Clustering
(2024-11)
He, Yanfang
The rapid advancement in hardware technology has generated a substantial volume of multi-view data with diverse representation formats. However, in practical applications, the collected multi-view data is often affected by noise due to various factors in the natural environment, making it challenging to obtain a high-quality dataset. To address the noise problem in multi-view data, this study enhances the gbs method and develops a new self-weighted graph multi-view clustering algorithm (swmcan). Particularly, swmcan addresses multi-view data noise using the l1-norm and optimizes the objective function through a novel iterative reweighted method. Extensive experiments on synthetic and real-world datasets consistently demonstrate that the swmcan algorithm outperforms recently proposed multi-viewclustering methods regarding clustering performance and noise robustness. Although the swmcan algorithm solves the noise problem in multi-view data, its initial and final graphs are independent and cannot learn from each other. To address this issue, this study incorporated joint graph learning from the gmc algorithm into swmcan, creating a new algorithm called swmcan-jg. The swmcan-jg algorithm effectively tackles both noise and independence problems simultaneously.

Browse

Recent Submissions