ANOMALY DETECTION IN IOT SENSOR DATA USING MACHINE LEARNING FOR REAL-TIME MONITORING SYSTEMS

Abstract

The high rate of growth of Internet of Things (IoT) devices in the industrial, healthcare, and smart city infrastructure sectors has produced new amounts of sensor data that need to be constantly analyzed in real-time. The overall challenge of identifying anomalies in these high-velocity, heterogeneous data streams is basic with important consequences of operational safety, system dependability and predictive maintenance. This paper is a systematic review of machine learning-based anomaly detection systems in IoT sensor setups. A systematic literature search was done in the IEEE Xplore, ACM Digital Library, ScienceDirect and Google Scholar, to identify publications published between 2014 and 2024, and a final corpus of 47 primary studies were identified using specific inclusion and exclusion criteria. The review is conducted on the statistical models, classical machine learning models, and deep learning architectures, and federated learning frameworks and their performance is assessed on the standard benchmark datasets, such as SWaT, SMD, SMAP, and MSL. Specific applications of industrial IoT, healthcare monitoring, and smart city systems are discussed in separate sections. The challenges in deployment such as constraints on real-time latency, concept drift, class imbalance, and integration of edge computing are discussed. The results show that both hybrid and federated deep learning models provide the most promising direction to scalable, privacy-preserving, and low-latency anomaly detection in commercial IoT systems, although interpretability and adversarial robustness are still open research problems.