Recent Advances in Anomaly Identification for IoT Devices Using Machine and Deep Learning Models

Abstract

The rapid expansion of the IoT has revolutionized data interchange in numerous industries, including smart cities, transportation, healthcare, and industry. However, it has also posed serious security problems due to the increasing number of vulnerable devices. Anomaly detection has become the primary instrument of the shield to locate abnormal behaviors in the box of the intrusion, fault, and failure of the operation. Anomaly detection in IoT systems has been the focus of recent advances in ML and DL methods. This paper assesses the efficacy, accuracy, and adaptability of supervised, unsupervised, and hybrid learning models in this context. In addition, the review emphasizes the main factors that determine detection performance, such as feature extraction, real-time detection, and multi-target learning strategies. Although ML/DL-based approaches have a strong potential, problems, for example, computational overhead, scalability limitations, and a lack of sufficient labeled data, that hinder the progress still exist. The paper provides useful information for researchers and practitioners in the choice of the most effective methods and assists in making the first steps toward future advances in IoT anomaly detection securing that it is still smart and resource-efficient