A Hybrid Approach for Geospatial Anomaly Detection in Sensor Measurements

Abstract

Sensor networks play a critical role in applications such as smart cities, industrial automation, and environmental monitoring. However, their reliability can be compromised by sensor faults, measurement inaccuracies, or environmental disruptions that lead to anomalous data. This study proposes a hybrid anomaly detection framework that integrates machine learning and spatial analysis techniques to improve the detection of such anomalies. Specifically, the approach leverages Isolation Forest and DBSCAN for anomaly detection, complemented by Moran's I index to capture spatial dependencies often overlooked by traditional methods. This combination enables the identification of both local and spatially clustered anomalies, addressing the challenge of geospatial irregularities in sensor measurements. Experimental validation across real-world datasets, including environmental, seismic, and meteorological sensor data, demonstrates that the proposed model outperforms conventional statistical methods in accuracy, recall, and robustness. Overall, this hybrid model offers a promising solution for enhancing the precision, reliability, and situational awareness of sensor-based monitoring systems.