HYBRID METHOD FOR ANDROID MALWARE DETECTION USING HYPERGRAPH NEURAL NETWORKS AND FEATURE OPTIMIZATION

Abstract

The article considers the problem of detecting complex and obfuscated malware on the Android platform. The evolution of protection methods is analyzed: from classical machine learning to deep neural networks. It is found that traditional methods often lose contextual connections between program components. A new approach is proposed, which is based on the construction of hyper graphs to model multidimensional dependencies between API calls and uses the bio-inspired Cheetah Optimization algorithm to select the most significant features. This approach allows to increase the accuracy of detecting malware families and reduce the overall load.