AI-Powered Anomaly Detection in Environmental Monitoring Systems Using Distributed Sensor Networks

Environmental monitoring systems are essential for supporting the sustainable and reliable functioning of IoT-enabled infrastructures. The data collected/generated by these sensors often affected by sensor malfunctions, environmental disturbances, or communication issues. Incorrect sensor data can lead to non-compliance with standards, risking penalties or legal issues. Therefore, detecting and removing anomalies from sensor data is critical for ensuring the accuracy, reliability, and safety of any data-driven system. This paper presents an AI-driven parallel anomaly detection framework. This framework integrates multiple anomaly detection algorithms - Isolation Forest, Local Outlier Factor (LOF), Autoencoder, and DBSCAN in parallel to generate individual anomaly scores or labels. These scores are passed through multiple postprocessing steps: anomaly score normalization, Temporal consistency check, ensemble decision mechanism, false positive reduction, etc. to finally detect anomaly accurately. This presented work has been evaluated using Intel Berkeley Lab Sensor Dataset, and Beijing PRSA Air Quality Dataset. The dataset includes temperature, humidity, light, and voltage readings from distributed sensors. The system achieved an anomaly detection rate of about 5.