Unspecific Event Detection in Wireless Sensor Networks

Event detection has always been an important application in the practical deployment of wireless sensor networks (WSNs). Although threshold based event detection method works well in some cases when significant characteristics exist for specific event, it has some limitations in many other application scenarios when we do not know much priori knowledge on incoming event. Inspired by the latest development of time series data mining technology, a flexible approach is proposed for unspecific event discovery in this paper. Event is defined as a pattern that is infrequent appeared. An algorithm is developed to find event patterns efficiently. We show how this approach outperforms threshold based method in Castalia simulation environment.