Classification of Malware Images Using Fine-Tunned ViT

Abstract

Identifying and classifying malware has become a critical task in ensuring the security and resilience of computer systems and networks. Traditional techniques for malware assessment often rely on signature-based methods, which struggle to keep up with the constantly evolving landscape of malware variations. Recently, the application of advanced deep learning methods has shown promising results in automating the malware classification process. This study presents an innovative strategy for classifying malware images using the Vision Transformer (ViT) architecture. The ViT model is adapted to the domain of malware analysis by representing malware images as input tokens. A comprehensive dataset of 14,226 malware samples from 26 families was used to evaluate the effectiveness of this approach. A comparative analysis was performed between the performance of the ViT-based classifier, traditional machine learning approaches and other deep learning architectures. Our experimental results demonstrate the potential of ViT in handling malware images, achieving a classification accuracy of 98.80%. The presented approach establishes a strong foundation for further research in utilizing cutting-edge deep learning architectures for enhanced malware analysis and detection techniques. © 2024, Sakarya University. All rights reserved.