Online data-centric anomaly detection framework for sensor network deployments

In this paper, we propose an online practical anomaly detection framework rooted in machine learning to identify data-centric anomalies in sensor network deployments. The framework enables application administrators to train a network of deployed sensors, instructs the nodes to extract online statistical features, and allows every node in the network to carry out the anomaly detection. Through simulation and a real-world in-door experimental deployment, our detection framework is shown to be able to identify data-centric anomalies with a very high accuracy (98% to 100%) while at the same time incurring much less memory, computation, and communication overhead compared to the state-of-the-art.