Real-Time Anomaly Detection in IoT Networks Using Deep Learning over Wireless Channels

Abstract

This study introduces IoT-AnomalyNet, a novel deep-learning approach designed for real-time anomaly detection in IoT networks operating over wireless channels. IoT-AnomalyNet combines long short-term memory networks (LSTMs), Convolutional Neural Networks (CNNs), autoencoders, attention mechanisms, and hybrid architectures to effectively identify patterns in both spatial and temporal dimensions within IoT sensor data streams. Through comprehensive experimentation with diverse datasets and IoT sensor readings, IoT-AnomalyNet achieves an impressive accuracy rate of 95.73% for anomaly detection. Notably, IoT-AnomalyNet outperforms traditional machine learning methods with remarkable recall (97.5%) and precision (95.5%) rates for normal instances and recall (97.85%) and precision (96.24%) rates for attack instances. These results underscore the efficacy of deep learning methodologies in accurately detecting anomalies in real-time IoT data streams transmitted via wireless networks. By proactively identifying abnormal behaviors, IoT-AnomalyNet holds significant promise in mitigating risks, ensuring continuous operation, and enhancing the security and reliability of IoT systems.