A SURVEY ON IDENTIFICATION OF ANOMALIES IN IOT SYSTEMS USING AI-DRIVEN MACHINE LEARNING AND DEEP LEARNING METHODS

Abstract

The Internet of Things' (IoT) explosive growth has brought previously unheard-of connections to industry, healthcare, smart cities, and smart households. However, the complexity and heterogeneity of IoT environments make them vulnerable to anomalies arising from faults, environmental changes, or malicious activities. For systems to be reliable, secure, and operate well, anomaly detection is essential.  In addition to DL architectures like CNNs, RNNs, LSTMs, GRUs, autoencoders, and GANs, supervised, unsupervised, and semi-supervised methods for anomaly detection in Internet of Things (IoT) systems are the main topics of this study's thorough examination of deep learning (DL) and machine learning (ML) models.   To improve detection accuracy, reduce false positives, and adapt to shifting assault patterns, it also examines hybrid ML–DL models, which include the best aspects of both theories.  This includes discussion of difficulties, including lack of resources, few datasets, interpretability of models, and resistance to hostile attacks. The paper also outlines real-world applications, comparative model analysis, and future research directions, emphasizing lightweight, privacy-preserving, and Scalable methods for anomaly detection in dynamic IoT contexts. Future work will explore integration with 5G, edge computing, and blockchain to enhance adaptability and real-time performance