Malware Detection Using Anomaly Detection Techniques: A Performance-Driven Approach

Abstract

With evolving cyber threats, traditional malware detection struggles to adapt to emerging variants. This study explores anomaly detection as an innovative approach to identifying malware by detecting deviations from normal system behaviour. Using a publicly available dataset, we evaluated models such as K-Means, Gaussian Mixture Models, Autoencoders, and Generative Adversarial Networks (GANs). The results show that deep learning models, especially GANs and Transformer-based models, achieve high accuracy. Our approach reduces false positives while ensuring computational efficiency for real-time deployment on resource-constrained devices.