Malware Detection Using Transfer Learning with ResNet-50: An Image-Based Approach
Mohammed Hameed, Malak Korami, Abdulahakim Al-Wahabi, Amr Al- Wajih, Mohammed Rageh, Ahmed Waleed, Roqaiah Mokhtar · 2025
Recent developments in the field of cybersecurity and deep learning have introduced new and innovative approaches to malware classification. Older malware detection methods struggle to keep up with modern threats due to advanced obfuscation and packaging techniques. In this study, we use a publicly available dataset where PE (Portable Executable Files) files were pre-converted to RGB images, allowing us to leverage deep learning techniques for malware classification. The proposed method relies on a pre-trained ResNet-50 convolutional neural network (CNN) to extract high-level features and classify malware families. In our experiments, the pre-trained ResNet-50 network was adapted to image-based malware detection and recognition using transfer learning techniques. The dataset called the Blended Malware Image Dataset is used, which holds a huge variety of malware families. Experiments result shows that our study unravels state-of-the-art classification accuracy and surpasses in classification performance with traditional machine learning models employed in the original study. These results demonstrate that the potential of image-based malware classification by deep learning may be a step toward stronger/automated cybersecurity solutions.