AI-driven approach for identifying abnormal behaviours in Internet of Things devices

The rapid expansion of Internet of Things (IoT) devices across sectors such as healthcare, manufacturing, and smart cities has led to a significant increase in data volume and complexity. Anomalous detection highlights the necessity for robust anomaly detection systems capable of identifying critical issues, including system malfunctions, security breaches, and operational inefficiencies. Traditional anomaly detection methods often struggle to handle the highly dynamic, high-dimensional nature of IoT data. This research introduces an artificial intelligence-based approach that integrates deep autoencoders with transfer learning to improve anomaly detection in IoT device behaviours. Deep autoencoders are used to learn detailed representations of normal operational patterns, enabling the identification of abnormal behaviour via significant reconstruction errors. To overcome challenges such as the diversity of IoT environments and the scarcity of labelled anomaly data, a transfer learning mechanism is adopted to transfer knowledge from data-rich domains to domains with limited labelled samples. This process enhances model generalization and reduces reliance on extensive labelling. The proposed method was evaluated on the N-BaIoT dataset and demonstrated superior performance compared to conventional methods, achieving Precision, Recall, and F-measure values exceeding ۹۰%. These findings indicate that combining deep autoencoders with transfer learning provides a scalable, adaptive, and highly accurate framework tailored to the unique characteristics of diverse IoT domains.

کلیدواژه ها:

autoencoders ، transfer learning ، Internet of Things ، anomaly detection