Malware Detection Based on Multi-level and Dynamic Multi-feature Using Ensemble Learning at Hypervisor
Jian Zhang,Cheng Gao,Liangyi Gong,Zhaojun Gu,Dapeng Man,Wu Yang,Wenzhen Li
DOI: https://doi.org/10.1007/s11036-019-01503-4
2020-01-08
Mobile Networks and Applications
Abstract:As more and more applications migrate to clouds, the type and amount of malware attack against virtualized environments are increasing, which is a key factor that restricts the widespread deployment and application of cloud platforms. Traditional in-VM-based security software is not effective against malware attacks, as the security software itself becomes the target of malware attacks and can easily be tampered with or even subverted. In this paper, we propose a new malware detection method to improve virtual machine security performance and ensure the security of the entire cloud platform. This paper uses the virtual machine introspection(VMI) combined with the memory forensics analysis(MFA) technology to extract multiple types of dynamic features from the virtual machine memory, the hypervisor layer and the hardware layer. Furthermore, this paper proposes an adaptive feature selection method. By combining three different search strategies, three types of features are compared and analyzed from three aspects: effectiveness, system load and security. By adjusting the weight of each feature, it meets the detection requirements of different malware in the cloud environment as expected. Finally, the detection method improves the detection accuracy and generalization ability of the overall classifier using the AdaBoost ensemble learning method with Voting's combination strategy. The experiment used a large number of real malicious samples, and achieved an accuracy of 0.999 (AUC), with a maximum performance overhead of 5.6%.
computer science, information systems,telecommunications, hardware & architecture