Homology Detection of Malicious Codes Based on a Fuzzy Graph Neural Network

Malicious code detection and homology analysis has been a hot research field in malicious code analysis. The API call graph extracted from malicious code can express the behavior information of malicious code effectively, but the efficiency of malicious code analysis based on graph structure is low because of the high algorithm complexity of solving subgraph isomorphism. To solve these problems, a new method of malicious code homology detection based on a fuzzy graph neural network is proposed. First, the malicious code features are selected and summarized, then the malicious code features are extracted, and finally the malicious code features are compared. Static analysis is used to analyze malicious code, extract its API call graph, and then process the API call graph to make it match the input of the convolutional neural network. With the help of the online code analysis platform, the platform adopts a sandbox mechanism, integrates more than 90 kinds of antivirus software, and analyzes the samples uploaded by users, which can give malicious evaluations. Experimental results show that the accuracy rate of malicious code homology analysis is 93%, and the accuracy rate of malicious code detection is 96%.