Ranking-based feature selection for anomaly detection in sensor networks

Anomaly detection, for uncovering faults and failures, is a crucial task for wireless sensor networks (WSNs). There have been substantive research efforts in this field such as source-level troubleshooting, rule-based inference, and time sequence event analysis. Most existing approaches, however, rely on the collection of a large amount of information. Due to the lack of management on information features, the redundancy of collected information greatly degrades the efficiency of diagnosis in large-scale WSNs. To address this issue, we propose RFS (Ranking-based Feature Selection), a three-stage approach to efficiently select representative feature sets for diagnostic tasks and effectively characterize the network status. RFS is a compatible component that can be integrated with most state-of-the-art diagnostic approaches. We conduct extensive experiments based on a large-scale outdoor WSN system, GreenOrbs, to examine the performance of RFS. The results demonstrate that RFS achieves effective anomaly detection in a largescale WSN with low overhead.