Explainable AI-Based Malware Detection Using Static Analysis and Machine Learning
Supervisor Name
Eman Daraghma
Supervisor Email
eman.yasser85@gmail.com
University
Palestine Technical University Kadoorie
Research field
Computer Science
Bio
Prof. Dr. Eman Yaser Daraghma is a Professor at Palestine Technical University–Kadoorie (PTUK), Palestine. She holds a Ph.D. in Computer Science and Engineering from National Chiao Tung University, Taiwan. Her research interests include Artificial Intelligence, Machine Learning, Cybersecurity, Digital Forensics, and Data Mining. She has extensive experience in teaching, research, academic leadership, and international research collaboration, including serving as Dean of the Faculty of Integrated Education at PTUK. She is actively involved in supervising research projects, reviewing scientific publications, and contributing to international academic initiatives.
Description
This project aims to develop and evaluate an explainable machine learning-based malware detection system for Windows executable (EXE) files. The system will use static analysis to extract Portable Executable (PE) features and machine learning algorithms to classify files as benign or potentially malicious without executing them. Using the EMBER malware dataset, the project will investigate and compare classification models such as Random Forest, XGBoost, and LightGBM. Particular attention will be given to feature effectiveness, detection accuracy, false-positive rates, and computational efficiency. Explainable AI techniques, such as SHAP, will be investigated to identify the features influencing classification decisions. The trained model will be integrated into a web-based application that allows users to upload EXE files securely, obtain classification results, and view relevant explanations. The system will be evaluated using precision, recall, F1-score, confusion matrices, and processing time. The project aims to contribute a practical and interpretable malware detection prototype while investigating the trade-offs between classification performance, static feature selection, and explainability.