Adaptive A IoT Security Framework for Real-time Anomaly and Intrusion Detection in Connected Devices

Uncategorized

Authors: Nafisa S, Dr. Balaji.K

Abstract: The emergence of AI and IoT (AIoT) technology fusion has opened up a new avenue for the creation of highly insecure conditions in terms of protecting continuous streams of information generated by the heterogeneous population of connected devices from highly advanced attacks. Existing static approaches suffer from limitations related to concept drift, limited resources and inherent dynamics of threat space. We introduce Adaptive AIoT Security Framework (AASF) combining lightweight edge-based anomaly detection with federated collaborative learning and online ensemble adaptation techniques. The proposed solution is built on the basis of two-level architecture: a resource-optimized sparse autoencoder used for computing local anomaly scores on resource-constrained devices, and federated aggregation layer hosted in the cloud that helps to keep model synchronization at regular intervals without revealing telemetry data. An adaptive thresholding technique relying on reconstruction error distribution allows an individual edge device to set up its sensitivity level for detection. Experimental results show that AASF outperforms existing solutions with respect to the detection performance while ensuring practical applicability on edge devices, as well as minimizing the amount of required communication overheads through selective updating.

DOI: https://doi.org/10.5281/zenodo.24414017

× How can I help you?