Abnormal-node Detection Based on Spatio-temporal and Multivariate-attribute Correlation in Wireless Sensor Networks

In wireless sensor networks (WSNs), data can be subject to malicious attacks and failures, leading to unreliability. This vulnerability poses a challenge to environmental monitoring applications by creating false alarms. To guarantee a trustworthy system, we therefore need to detect abnormal nodes. In this paper, we propose a new framework for detecting abnormal nodes in clustered heterogeneous WSNs. It makes use of observed spatiotemporal (ST) and multivariate-attribute (MVA) sensor correlations, while considering the background knowledge of the monitored environment. Based on the ST correlations, the collected data is analyzed by computing the crosscorrelation between sensor streams. A new method is proposed for evaluating the intensity of the correlation between two sensor streams. The crosscorrelation value obtained is compared against two thresholds, the lag threshold and the correlation threshold. Based on available background knowledge and the observed MVA correlations, a number of rules are presented to detect abnormal nodes while identifying real events. Our experiments on real-world sensor data demonstrate that our approach captures the correlation and discovers abnormal nodes efficiently.