A Deep Learning Approach for Robust Internet Malware Detection via Eigen Space Analysis

Abstract

The Internet of Things (IoT), particularly in military applications, comprises a wide array of connected devices such as medical systems and smart combat gear, making them prime targets for sophisticated cyberattacks. A common attack strategy is the deployment of malware, which can compromise device integrity and mission-critical operations. In this study, we introduce a robust malware detection approach tailored for internet of battlefield things (IoBT) environments using deep Eigen space learning. The method involves transforming device operational codes (Opcodes) into a vector space and applying deep learning to accurately classify benign and malicious applications. To counter anti-forensic techniques like junk code injection, an affinity-based feature selection technique is employed to filter non-instructive Opcodes, improving detection performance and resilience. Experimental results confirm the robustness of our model against various evasion tactics, including repeated code injection. A dataset of malware samples used in our analysis has been released publicly to aid future research in this domain. Our findings demonstrate a significant advancement in the secure deployment of IoT systems in sensitive, high-risk contexts.