An energy-efficient anomaly detection approach for wireless sensor networks

Detecting anomalies in sensed data in a wireless sensor network (WSN) is essential to identify malfunctioning nodes. A key challenge is to develop an approach to dynamically detect anomalies with an acceptable accuracy while the communication overhead and energy consumption in the network is minimized. In this paper, we propose an energy-efficient approach for detecting anomalies in sensed data in a WSN. The anomaly detection procedure is comprised of two phases: training and detection. In the training phase, every sensor node computes a description of its local sensed data and sends it to its group head. After receiving descriptive data from member nodes, the group head applies approximate global principal component analysis (AGPCA) to establish a global normal profile and sends it to all member nodes. In the detection phase, every member node detects anomalies based on their projection distances from the global first principal component. As environmental conditions change over time, the established global normal profile is updated using weighted coefficients. We demonstrate that the proposed approach achieves high detection rate with low false alarm rate while minimizing the energy consumption in the network.