Visualization Feature and CNN Based Homology Classification of Malicious Code

Abstract

The malicious code brings a serious security threat. Researchers have found that many new types of malicious code are variants of the existing one. The homology classification of the unknown malicious code can find its corresponding family in which all the code share inherent similarities from the database, so that the defenders can make rapid response and processing. We use the algorithm of malicious code visualization to translate the homology classification problem into the image classification problem. A convolution neural network for malicious code image is constructed. We train it to complete the malicious code homology classification on two different datasets. The results show that our work outperforms most of existing work with the accuracy of 98.60%.