Evaluating machine learning-driven intrusion detection systems in IoT: Performance and energy consumption

Abstract

In the landscape of network security, the integration of Machine Learning (ML)-based Intrusion Detection System (IDS) represents a significant leap forward, especially in the domain of the Internet of Things (IoT) and Software-Defined Networking (SDN). Such ML-based IDS are crucial for improving security infrastructures, and their importance is increasingly pronounced in IoT systems. However, despite the rapid advancement of ML-based IDS, there remains a gap in understanding their impact on critical performance metrics (e.g., CPU load, energy consumption, and CPU usage) in resource-constrained IoT devices. This becomes especially crucial in scenarios involving real-time cyber threats that challenge IoT devices in a public/private network.

To address this gap, this article presents an empirical study that evaluates the impact of state-of-the-art ML-based IDSs on performance metrics such as CPU usage, energy consumption, and CPU load in the absence and presence of real-time cyber threats, with a specific focus on their deployment at the edge of IoT infrastructures. We also incorporate SDN to evaluate the comparative performance of ML-based IDSs with and without SDN. To do so, we focus on the impact of both SDN’s centralized control and dynamic resource management on the performance metrics of an IoT system. Finally, we analyze our findings using statistical analysis using the Analysis of Variance (ANOVA) analysis. Our findings demonstrate that traditional ML-based IDS, when implemented at the edge gateway with and without SDN architecture, significantly affects performance metrics against cyber threats compared to DL-based ones. Also, we observed substantial increases in energy consumption, CPU usage, and CPU load during real-time cyber threat scenarios at the edge, underscoring the resource-intensive nature of these systems. This research fills the existing knowledge void and delivers essential insights into the operational dynamics of ML-based IDS at edge gateway in IoT systems.