Thesis (Bachelor of Science in Computer Science)
Browse
132 results
Search Results
Item Culturally adaptive neural network for detecting cybersecurity vulnerabilities in Bangladeshi web applications(BRAC University, 2026) Shaolin, Mohosina; Nawar, Fariha; Siddique, Arik Ahmed; Shams, Shaikh Mohammad Ali; Maliyat, Nafisa; Mostakim, Moin.As cyber threats become more complex and frequent, conventional methods for detecting website vulnerabilities, such as rule-based and heuristic approaches, faces significant difficulties, including limited adaptability, high rates of false positives, and a lack of contextual insight. This study presents a predictive model based on neural networks aimed to actively evaluating website security. By applying essential features like security headers, SSL/TLS settings, and SQL injection vulnerabilities, the model detects complex patterns and irregularities, enabling precise identification of emerging threats and vulnerabilities. This approach uses data-driven feature engineering and training with custom neural architectures, for comparison we used random forest and gradient boosting, For explainability we used SHAP followed by evaluation metrics such as precision, recall, and F1-score. Key results show improved accuracy, reduction of false positives, automated monitoring of configurations, and enhancement of resilience against adversarial attacks. Although neural networks show significant potential for transformation, challenges related to transparency, computational demands, and data imbalance are acknowledged. This highlights the necessity for ongoing learning, scalability, and integration with current frameworks, laying the groundwork for robust and adaptable web security strategies.Item Temporal state-aware unsupervised anomaly detection for industrial control system(BRAC University, 2026-01) Anik, Khandoker Wahiduzzaman; Ontu, Md. Rakib Hossain; Sohag, Md. Mehedi Hasan; Badhon, Fardin Jahan; Chakrabarty, AmitabhaThe increasing interrelationship with Information Technology infrastructure between the Industrial Control Systems (ICS) and critical infrastructure has presented advanced cyber-attacks to critical infrastructure, which not only places data security at risk but also threatens the physical safety, to operational continuity. The thesis will visit the issue of creating an efficient, interpretable and computationally efficient intrusion detection system in an ICS environment through a proposed novel LSTM auto-encoder architecture, specifically trained with edge deployment in mind. The article takes a rigorous approach where physics-conscious feature engineering is embraced, deep-learning architecture creation, and thorough assessment of the WADI (Water Distribution) benchmark data. The proposed system achieves a score of 0.7018 in F1 (Precision=0.7196, Recall=0.7149) by performing the dimensionality reduction of 127 sensors to 30 (which is a reduction of 76 per cent), and by adding the zero-crossing-rate features to the frequency-domain analysis, which is drastically higher than more traditional statistical methods, including Isolation Forest (0.58 F1), or the current state-of-the-art methods, including STADN. To be practical, the system is edge-compatible with an inference latency of 1.84ms, a million parameters, and consumes 5.38W of power when run on simulated NVIDIA Jetson Nano hardware, which is a ten-fold faster inference time than graph-based algorithms. Unsupervised approach, which learns only based on normal operational data, helps to detect novel, zero-day attacks, therefore overcoming the limitation of labelled attack data in operational settings. The study establishes that advanced deep-learning systems can be deployed on the tight computational requirements of industrial edge devices, thus creating a reproducible model of secure and real-time secure critical infrastructure protection.Item Using machine learning to predict optimal erasure coding policies for object storage system in OpenStack Swift(BRAC University, 2026-01) Ankon, Amio Malakar; Chaki, Boloy; Sayan, Mashrur Shakhawat; Sarker, Debashish; Mukta, Jannatun NoorErasure coding helps to reduce storage overhead and improve fault tolerance. But the procedure to select an appropriate erasure coding policy is complex, which often involves tradeoffs among various metrics such as access latency, recovery behavior, storage efficiency, etc. Generally, these are handled using static or heuristic-based configurations, which can not account for variations in workload. In the industry, service providers like Ceph do benchmarking based on throughput/latency without considering workload diversity. OpenStack Swift, one of the most widely used open-source object storage systems, supports erasure coding, but it allocates policy selection in a manual, static way - without it being workload-aware. To address this problem, this thesis showcases a data-driven performance modeling framework for erasure coding in object storage systems using machine learning. A structured dataset- ABDS-30k was constructed in a controlled execution of varying workload conditions. It was created on a Swift All-In-One (SAIO) testbed under different erasure coding policies with the addition of failure injections. The dataset collects several empirically observed performance metrics such as read and write latency, tail latency, success rate, and reconstruction time in case of disk failures. For the job of selecting the optimal erasure coding policy, we adopt two distinct Machine Learning paradigms - a regression-based approach of performance modeling and a classification-based approach for direct data-driven policy recommendation. For regression, CatBoost, XGBoost, and Random Forest Regression are used to predict target metrics in a given workload context and failure scenario, and the optimal policy is chosen based on a weighted score. In the classification-based approach, CatBoostClassifier, XGBoostClassifier, and Logistic Regression are used to label workload–policy pairs using an oracle cost function, and the most optimal erasure coding policy is directly recommended. Experimental results show that the regression models have moderate prediction errors for latency and recovery metrics because of the inherently noisy and heavy-tailed nature of the system, but they remain effective in optimal policy recommendation by exhibiting top-1 accuracy with 48.16% in XGBoost Regression and a top-3 accuracy 100% in Random Forest Regression model. Also, the regret mean (0.010 - 1.81) and regret median (0.00216 - 0.089) values showed a very low margin of error. Classification-based approach shows comparatively weaker metrics, indicating that it is not optimal for data-driven policy recommendation. Overall, this shows the feasibility of using ML-based performance modeling as opposed to static and heuristic-based policy selection, which can be later used as a foundation for future control-plane automations.Item Adaptive traffic signal control for urban intersections using reinforcement learning: a SUMO simulation case study on Gulshan-2(BRAC University, 2025-02) Shamim, MD Shahadat Hossain; Chowdhury, Tawhid; Deep, Sadid Arman; Bhuban, Riazul Hoque; Chakrabarty, AmitabhaTraffic congestion is an ever-growing concern in rapidly increasing population and increase of vehicles on roads. On top of that, our road condition, pedestrian be- havior, and driving behavior makes the situation much worse. As Bangladesh has started adapting to the fixed-time traffic light system, finding the optimal adaptive traffic light management system is required to mitigate the issue. Keeping the lack of intra-vehicular communication devices in mind, we have studied different traffic light controlling systems and how these perform on Bangladeshi roadways. In our study, we have used traffic simulator to simulate Gulshan-2 intersection, one of the major congestion points in Dhaka city. Based on traffic data, we have created a vehicular network system to find out how the existing traffic light models per- form on mitigating the traffic jam using an adaptive traffic management system. In our study, we simulate the Gulshan-2 intersection using a realistic traffic dataset de- rived from field data and analyze three control models, Fixed-Time, Q-Learning, and Deep Q-Network (DQN). The results show that in ideal traffic condition, when no anomalies are present; the Q-Learning controller reduced average vehicle delay by approximately 60-70% and improved throughput by 35-45% compared to the fixed- time system, demonstrating its superior adaptability to dynamic traffic flows. How- ever, when we add real-life anomalies such as potholes and jaywalking, the baseline controllers’ performance drops. By acknowledging the drawbacks of QL controller in abnormal scenarios, we have introduced a hybrid model CRQL (Congestion Re- sponsive Queue Learning) and made some improvements. In abnormal situation CRQL performs better in regard to throughput, which is a 76.9% improvement over Fixed-Time, 37.4% improvement over Q-Learning and 215.2% improvement over DQN. This reflects a overall performance improvement under disruptive and non- stationary environment. In short, the study helps to see how existing models perform under real-life anoma- lies, which factors are significant to create Ad hoc model for our extreme pedestrian patterns and road conditions and how our proposed solution; the CRQL ad hoc or hybrid model performs to address the problems.Item Machine learning based multi-variate MBB-user growth prediction and worst-cell clustering in cellular network(BRAC University, 2024-02) Ahmed, Tanjir; Arefin, Mohammad ShamsulThe thesis will analyze mean user number of cellular networks based on maximum user, data volume and cell availability from a KPI report generated from OSS server. Firstly, we will collect KPI data from OSS server of specific area where large number of cellular users taking service from network and then conduct data cleansing and data splitting. Finally, we will analyze the data with multiple variable regression and support vector machine (SVM) analysis to predict user number. We will also perform K-means clustering method to find the worst cells to improve the network performance and user experience. We will do the analysis with most popular and latest python which boasts the feature of pure object-oriented programming, platform independent, concise and very elegant language. So, we will call the corresponding library function to predict the user number and worst cells clustering which will help a telecom network operator to plan, design an effective and optimized network and also help to improve user experience as well. Moreover, this analysis will help to make cluster plan, and understand user behavior in a specific area.Item Energy cost minimisation to support electric vehicle charging using machine learning(BRAC University, 2025-12) Fahad, Mohammed Amir; Ahmed, Mst.Sanjida; Niha, Ethika Adrita; Shaon, Sabbir Hossain; Abdullah, Samia; Chakrabarty, AmitabhaThe rapid growth of EVs and the integration of renewable energy has made the administration of charging demand and reduction of electric expenses more complex. This paper introduces an dual machine learning forecasting of EV charging demand, solar photovoltaic (PV) energy generation, and PV-first smart charging scheduling and cost calculation.The realistic daily energy profiles were developed using real data on fifty EVs provided in the workplace. In the case of EV demand forecasting five deep learning models were tested, LSTM, GRU, TCN, Transformer Encoder and a Hybrid LightGBM-LSTM model where our proposed hybrid model had the lowest RMSE value 0.4049 and R2 value 0.8608. In the case of PV generation forecasting, eight ML models were experimented in the form of Random Forest, Decision Tree, Gradient Boosting, Linear Regression, LSTM, GRU, TCN as well as the hybrid model. Again, the highest PV prediction accuracy was obtained in Hybrid LightGBM + LSTM with RMSE value 273.55 and the R2 value 0.9321.The predicted EV and PV forecasts were combined in a PV- first scheduler that gave preference to solar power over grid power. In 7 days, the system made 41.87% cost savings and 52.3% contribution of solar; in 30 days, it made 17% savings under flat pricing and 24.9% savings under TOU that is aided by 37.7% renewable contribution. These findings indicate that dual hybrid forecasting and PV-first smart charging scheduling are effective in the reduction of charging expenses, minimization of grid reliance,minimize CO2 and in line with SDG 7 and SDG 13 on clean and sustainable energy.Item Machine learning-based approach to improving Parkinson’s disease diagnosis through voice signal analysis(BRAC University, 2025-10) Shanto, Arman Hossain; Suchi, Upoma Deb; Chowdhury, Raida Jafar; Mohammed, Afnan; Reza, Fairuz Suhala; Reza, Md TanzimParkinson’s disease (PD) is a chronic neurodegenerative disorder that influences motor and non-motor function and imposes huge health-related costs on society. Early and precise diagnosis still presents a big challenge because clinical examinations are subjective, and diagnostic methods based on imaging are expensive and invasive. As the vocal deficiencies are usually present in early PD, voice is considered to be a potential non-invasive and widely available neurobiological marker for early diagnosis. This thesis presents an end-to-end framework for PD detection via ML and XAI. Voice datasets were preprocessed for normalization, class balancing and dimensionality reduction in order to improve data quality. Acoustic Feature Selection We used a range of feature selection methods, including statistical significance testing, SelectKBest ranking, and principal component analysis to extract discriminative sub-sets of acoustic features. Various machine learning models were trained and tested, the ensemble methods performed well. The results indicate that the oversight model based on the selection of only 100 picked features with XGBoost approach provides high accuracy, precision and recall as well as F1-scores greater than 0.85 on the test set. SHAP and Morris Sensitivity Analysis were used to increase the interpretablity, revealing important indicators of PD-related vocal changes including jitter, shimmer and MFCC features. These observations were consistent with clinical data, which linked computational modeling to biological knowledge. The results indicate that voice coupled with interpretable ML techniques can be used as a robust digital biomarker for PD. This work will help build scalable, accessible and explainable tools for early diagnosis and remote healthcare.Item Intelligent sound localization and recognition with tactile feedback system to assist hearing-impaired individuals(BRAC University, 2025-10) Arony, Radia Tasnim; Jalil, Naomi Afrin; Chowdhury, Maisha Raidah; Tahsin, Umma Hafsa; Hasan, Mohammad Shabab Bin; Sadeque, Farig YousufA significant portion of the global population suffers to some extent from hearing loss and requires assistive devices. The everyday lives of hearing-impaired individuals are hampered by their inability to recognize important environmental sounds in their surroundings. This thesis proposes an innovative system that can both recognize and localize critical environmental sounds while providing tactile feedback to enhance user awareness in real time. Machine learning models such as Convolutional Neural Networks (CNNs) are utilized to classify specific sounds such as car horn, dog bark, scream, and calling bell. Furthermore, localization algorithms are integrated into a wearable device to determine the direction of these sounds. Once these sounds are detected and localized, the user is notified through vibration feedback from their smartphone or smartwatch, while the sound type, confidence level, and direction are simultaneously displayed. The proposed system is designed to be a low-cost, compact, and accessible solution for recognizing important environmental sounds and represents a meaningful advancement in the field of assistive devices by significantly enhancing the safety and quality of life of individuals with hearing impairments.Item Predictive modeling and simulation techniques for landslide risk management(BRAC University, 2025-06) Samit, Chowdhury Mohammad Mutamir; Wazed, Arian; Rasel, Annajiat AlimLandslides are common natural disasters in the hilly areas, inflicting significant damage to both the human lives and economy. Unlike other severe disasters like floods, earthquakes etc. landslides do noticeably have a significant impact on the development initiatives. Landslides are regulated by different triggering events, which makes it impossible to forecast their exact mechanism. During the last decade, researchers have focused on using machine learning to forecast landslides. The aim of our project is to estimate the probability of landslides. In our project we will use a set of 8 features to train the model and forecast landslides. The acquired data was studied by data count ,correlation matrix and distribution of feature data.We will be analyzing the biggest landslides and find the main reasons responsible behind these landslides. The dataset will be used to test various machine learning algorithms and examine a variety of factors and visualizations.We will then examine the models to determine which performs well and have better prediction accuracy.Item Early and late fusion ensemble methods for predicting bug severity from bug report(BRAC University, 2025-06) Farooq, Md. Farhan; Nabil, MD. Shahariar Nawshad; Azmain, Md. AquibAccurate prediction of software bug severity is essential for optimizing resource allocation, enhancing bug triaging, and improving project management within the software development lifecycle. This study introduces a robust methodology for predicting bug severity by leveraging textual data from bug reports, employing advanced natural language processing (NLP) techniques and machine learning models. We evaluate several approaches, including Word2Vec with XGBoost (68% accuracy, 64% precision, 68% recall), TF-IDF with Logistic Regression/SVM (77% F1 score), DistilBERT (73% accuracy, 70% F1 score), and DistilRoBERTa (76% accuracy, 73% F1 score), each demonstrating strengths in capturing semantic and contextual nuances of bug descriptions. To further improve performance, we propose a fusion-based ensemble learning framework, combining early fusion (integrating TFIDF, Word2Vec, and transformer embeddings into a unified feature vector) and late fusion (aggregating predictions from independently trained models). The hybrid Ensemble Fusion model achieves the highest performance, with an accuracy of 79% and an F1 score of 76%, excelling in generalizing across diverse bug severity, including challenging short and long durations. Our methodology encompasses rigorous data preprocessing, feature engineering, and techniques to mitigate class imbalance, utilizing a comprehensive dataset of bug reports with rich textual and metadata attributes. The results underscore the efficacy of integrating diverse feature representations and model predictions, providing a scalable, robust, and actionable solution for predicting bug severity, ultimately enhancing software development efficiency and reliability.