Malware detection approach based on deep convolutional neural networks

Abstract

Malware detection field becomes more valuable nowadays regarding the continuously growing number of malware codes emerging everyday. Besides, machine learning techniques have been widely used in various fields. For the purpose of employing machine learning in malware detection, an executable file should be represented by its features. Therefore, a dataset of labelled benign and malicious files is considered. Then, the developers extract the appropriate features to their model from each file. These features are displayed as inputs to a machine learning classifier. In previous researches, multiple features and classifiers were adopted in different combinations for a better classification. In this paper, we have been interested to PE header fields' features, and a deep convolutional neural network for classification. We extracted the bytes of the PE header fields' values and fed them to our model as greyscale images. Our model is constituted of 31 consecutive convolutional layers. The model was trained on the train dataset, and finally tested on the test dataset. The results were impressive reaching a test accuracy of 97.85%.