Malware detection based on directed multi-edge dataflow graph representation and convolutional neural network

Abstract

In recent years, malware has grown constantly in both quantity and complexity. Traditional malware detection methods such as string search, hash code comparison, etc. have to face the challenging appearance of more and more new malware variations. One of the most promising approaches to tackling them is to use machine learning techniques to automatically analyze and detect unknown malicious softwares. In this paper, we introduce a novel method of using dynamic behavior data to represent malicious code in the form of multi-edge directed quantitative data flow graphs and a deep learning technique to detect malicious code. Our experimental result shows that the proposed method archived a higher detection rate than other machine learning methods, and a higher unknown malware detection rate, compared with commercial antivirus software.