Fog Enabled Anomaly Detection System for Sensors’ Anomaly in IoT Environment Using Machine Learning

In the rapidly evolving Internet of Things (IoT) domain, artificial intelligence (AI) has transformed IoT systems into self-sufficient entities capable of making groundbreaking autonomous decisions. This paradigm shift has been bolstered by a remarkable increase in computational efficiency, allowing for AI integration, even within stringent resource limitations. However, implementing AI at the edges of such networks remains a significant financial challenge. IoT ecosystems comprising myriad devices, sensors, and actuators are inherently diverse in temporal and data profiles and communication protocols. This diversity predisposes these systems to a spectrum of anomalies. We introduce an innovative anomalydetection framework for sensor anomalies in IoT environments to address this critical challenge. Our methodology was rigorously evaluated using real-time temperature and humidity sensor data and a standard dataset provided by Schneider Electric. We leveraged four cutting-edge machine learning models— Logistic Regression, Random Forest, XGBoost, and AdaBoost—to gauge the framework’s efficacy. Furthermore, our system was tested against the established FogDLearner framework using a PureEdgeSim Simulator to replicate a fog-computing scenario. These results were compelling. The accuracy rates for the real-time sensor data were 98.00%, 98.65%, 99.00%, and 99.21% for Logistic Regression, Random Forest, XGBoost, and AdaBoost, respectively. For the Schneider Electric standard dataset, the models achieved accuracies of 99.00, 99.99, 99.98, and 99.99%, respectively. These findings underscore the robustness of our framework and highlight the potential of fog-enabled machine learning in revolutionizing anomaly detection in IoT environments.