A Multistage Malware Detection and Classification Model Using Visual Features

The development of the Internet and its widespread usage among users has led to an increase in malware activity in recent years. It has become increasingly challenging to identify and categorize malware samples using current anti-malware scanners as they have become capable of evading detection through obfuscation. A simple yet effective multi-stage model based on CNN is proposed for viewing, detecting, and classifying malware. The dataset is generated by combining Microsoft's Malware Classification Challenge dataset and a compilation of benign open-source code files. A visualization technique is proposed to convert different code files to images and the result exhibits an accuracy of 95% for detection and 94% for malware classification.