Battery-drain attacks pose a stealthy yet critical threat to Internet of Things (IoT) networks, enabling adversaries to exhaust device power without triggering traffic-based alerts. Existing intrusion detection systems (IDS) seldom address such energy-aware threats or report false alarm rates under realistic conditions. This paper introduces the Multi-View Energy-Aware Intrusion Detection System (MV-EA-IDS), a dual-view framework combining network-level analysis with inferred Energy Proxy Features to detect both conventional and power-drain attacks. It integrates a supervised Random Forest for traffic classification and an unsupervised Isolation Forest trained on benign energy proxies to capture anomalous drain patterns. Adaptive percentile thresholds and a meta-heuristic k-of-n fusion rule further minimize false positives. On the public CIC-IoT2023 dataset, MV-EA-IDS achieves 99.87% accuracy and only 0.01% false positives, establishing the first reproducible benchmark for energy-aware IoT intrusion detection.
Berchi et al. (Tue,) studied this question.