Probabilistic Anomaly Detection for Wireless Sensor Networks

Wireless Sensor Networks (WSN) are increasingly gaining popularity as a tool for environmental monitoring, however ensuring the reliability of their operation is riot trivial, and faulty sensors are not uncommon; moreover, the deployment environment may influence the correct functioning of a sensor node, which might thus be mistakenly classified as damaged. In this paper we propose a probabilistic algorithm to detect a faulty node considering its sensed data, and the surrounding environmental conditions. The algorithm was tested with a real dataset acquired in a work environment, characterized by the presence of actuators that also affect the actual trend of the monitored physical quantities.