A distributed anomaly detection model for wireless sensor networks based on the one-class principal component classifier

Abstract

The application of wireless sensor networks (WSN) is increasing with the emergence of the 'Internet of Things' concept. Nonetheless, the sensed data quality and reliability are sometimes affected by factors such as sensor's faults, intrusions and unusual events among others. Consequently, the real time and effective detection mechanisms of anomalous data are necessary for reliable decisions. In this paper, we proposed a one-class principal component classifier (OCPCC) based distributed anomaly detection model for WSN, which utilises the spatial correlations among sensed data in closed neighbourhoods. The feasibility of the model was validated using real world datasets and compared with local detection and some existing detection approaches from literature. The results show that the proposed model improves the detection rate of anomalous data compared to local model. A comparison with existing distributed models reveals the advantages of the proposed model in terms of efficiency while achieving better or comparable detection effectiveness.