AI-Driven Anomaly Detection and Behavior Analytics in Heterogeneous IoT Ecosystems

Abstract

With the growth of the IoT ecosystems, scalable, automated behaviour analytics becomes a major necessity. The paper provides an AI-based scheme in passive anomaly detection and behavioural profiling of heterogeneous IoT systems. The methodology is a blend of feature extraction on network flows and autoencoder-based unsupervised learning in order to identify subtle behavioural changes in devices. The model was tested on multi-protocol datasets (Wi-Fi, Zigbee and LoRaWAN) and recorded 94.6% detection and very few false positives. The findings show that smart passive surveillance can improve situational awareness and strengthen security in massive and multi-vendors infrastructures using IoT.