A Trustworthy and Robust Model for Mutated Malware Detection Via Explainable Ai and GANs
Truong Thi Hoang Hao, Truong Thi Hoang Hao, Nguyen Thai Duong, Nguyen Thai Duong, Nghi Hoang Khoa, Nghi Hoang Khoa, Van-Hau Pham, Van-Hau Pham, Phan The Duy, Phan The Duy · 2024
In recent years, researchers have continuously im-proved malware detection systems and successfully integrated machine learning (ML) and deep learning (DL) models into the detection and classification of malware samples. However, along with these improvements, adversarial attack techniques have also become more sophisticated. To combat the ever-evolving adversarial attack forms, many studies have been proposed and implemented. Adversarial training is one of the most common approaches. While it has been experimentally shown to mitigate some adversarial attacks, its effectiveness remains limited, hindering user confidence in its reliability. In this paper, we introduce RMMD, a robust model for detecting mutated malware, including packed and adversarial variants. This approach leverages Explainable AI (XAI) and Generative Adversarial Networks (GANs) in conjunction with adversarial training techniques. Our proposed model, RMMD, achieves a 90% accuracy rate across all models, indicating a significant leap forward in detecting packed and adversarial malware.