A hybrid model for classifying malware based on ResNet and Transformer
Kewei Li, Fudong Liu · 2023
Malware is becoming a significant hidden threat to network security, with traits like rapid population growth, numerous family varieties, and effective hiding. Traditional analysis methods have had trouble efficiently achieving family detection and classification in the face of numerous harmful code variants. This work investigates the structure and behavior analysis of malware using deep learning and suggests an analysis method that combines the RGB image features and behavior sequence features of malicious code. The classification of malware is made possible via image analysis and natural language processing. To achieve the extraction of image features and sequence features, the Transformer+ResNet network's hybrid malicious code classification model is applied. The accuracy rate of the model, which was tested on The Microsoft Malware Classification Challenge (BIG 2015) malware dataset is 99.32%, and its overall performance is satisfactory. It demonstrates the semantic complementarity of malware-related image features and sequence features in many dimensions and the successful classification of the malicious code family by the suggested model