Benchmarking Machine Learning based Intrusion Detection for IoT Edge Devices

Machine learning-based Intrusion detection (ML-IDS) gained a lot of interest from the research community as security threats facing the IoT are on the rise. Researchers focused on improving intrusion detection efficiency based on machine learning. However, the feasibility of applying proposed ML algorithms to resource limited IoT devices limited. We present a proof of concepts (POC) for ML-IDS for IoT edge device using actual devices. Furthermore, we envisage the lack of virtual environments that mimics the realistic IoT world, puts research completeness on bay. In this paper, we extend the benchmarks of intrusion detection, applied to IoT edge devices of a varying of computing resources. The benchmarks are based on machine learning techniques endorsed by the research community, and adopt popular intrusion datasets for training and testing models. We also utilize the benchmarks to calibrate Virtual-box environment in a novel way to produce virtual machines that closely mimic realistic IoT edge devices. Thus accurate virtualization would become a reliable tool available to researchers to test their approaches.