Enhancing IoT Security: Novel Mechanisms for Malware Detection using HPCs and Neural Networks

Abstract

With the exponential rise of internet-connected devices, the number of Internet of Things (IoT) devices has surpassed that of traditional IT devices. This proliferation in IoT adoption can be attributed to the growing demand for manufacturing automation and the desire for enhanced quality of life, leading to the production of smart devices in various industries. However, this rapid adoption has caught the attention of malicious actors, resulting in a significant increase in cyberattacks targeting IoT devices. In response to this emerging threat landscape, research on IoT security has been active. Nevertheless, the lack of commercial tools specifically designed for IoT device security raises concerns about the ability of security research and adoption to keep pace with the rising number of malicious actors. To address this gap, this study focuses on introducing a novel mechanism for detecting malware in IoT devices. By conducting experiments, we demonstrate that using Hardware Performance Counters (HPCs), complemented by physical features such as power consumption, can improve the current malware detection capabilities. Specifically, we employ Recurrent Neural Networks (RNN) and Multi-Layer Perception Neural Networks (MLP) to achieve a remarkable detection accuracy of 95% within a timeframe of less than 10 seconds from infection.