Simplified-Xception: A New Way to Speed Up Malicious Code Classification

Abstract

Traditional malicious code detection methods require a lot of manpower and resources, which makes the research of malicious code very difficult. The selection of malicious code features mainly relies on the subjective analysis and selection of experts, which has a large impact on the detection effect of the model. In this paper, malicious codes are converted into greyscale images as model inputs, and features are automatically extracted using a deep-learning model. An improved convolutional neural network model based on Xception (Simplified Xception) is proposed for malicious code family classification. The model reduces the number of modules in the original model and adds a depth-separable convolutional layer with a step size of 2 to enhance the generated grey-scale images. The model is compared with CNN models, ResNet50, and improved models related to Inception. The experimental results show that the accuracy of SimplifiedXception is 98%, which is better than other related models. Compared to the Xception model, the accuracy of the Simplified-Xception model was improved by 1.3% and the number of parameters was reduced by half.