Anomaly Localization by Joint Sparse PCA and Its Implementation in Sensor Network

Principal Component Analysis based anomaly detection approaches have been extensively studied recently. However, none of these approaches address the problem of anomaly localization. In this paper, we proposed a novel approach based on PCA to perform anomaly detection and localization in sensor network simultaneously. By enforcing the joint sparsity across the Principal Component in the abnormal subspace, we can accurately localize the abnormal sensor nodes from normal nodes. We demonstrate the localization performance in the experimental study using two real world data sets.