A Hybrid Deep Learning Model for Malicious Behavior Detection

Abstract

In recent years, malicious programs seriously threaten the security of information system. Because of its particularity, complexity and vulnerability, power information system is difficult to detect and kill by traditional anti-virus software. To solve the above problems, this paper proposes a malicious behavior detection method based on deep learning, which can identify malicious programs and attack types according to the activities of malicious programs and malicious software behaviors. In this paper, a hybrid deep learning structure based on convolutional neural network (CNN) and long-and-short term memory (LSTM) network is proposed. The call sequence of API is combined with other statistical features. The vector information obtained is input into LSTM unit through convolution, and the classification results are obtained through the above model. Compared with the existing methods, this method significantly improves the preparation rate and execution efficiency.