Anomaly detection through outlier and neighborhood data in Wireless Sensor Networks

Anomaly detection through finding outlier measurements is an important issue for monitoring application using a large databases gathered by Wireless Sensor Network (WSN) like medicine and military. In this paper, we evaluate a detection of outliers based on the distance between the current measurement and its neighbors. Our detailed evaluation supports a synthetic database generated with random values inserted into a real database from Intel Berkeley lab. In each round test, we increment the size of the learning window in order to have more outliers measurements. The study of results highlights the importance of accuracy of detection that its average is 89%. Moreover, the detector provides a low false alarm rate with an average of 10% and a sufficient detection rate that can reach 100%.