Dynamic Collaborative Change Point Detection in Wireless Sensor Networks

With wireless sensor networks (WSN) now readily available and capable of monitoring multiple physical phenomena over time, large volumes of data can now easily be generated in the form of multiple co-evolving data streams. This presents a number of challenging tasks for the analyst, who often seeks to monitor such data in real-time for the purposes of summarisation, anomaly detection and prediction. WSNs often suffer from severe resource constraints that prevent them from applying computational algorithms on large datasets as in conventional systems. Sensomax is an agent-based and object-oriented WSN middleware, which is capable of executing multiple concurrent applications based on their required operational paradigm. Its component-based architecture features seamless integration of light-weight computational algorithms at different levels throughout the network. This paper presents the preliminary work on a novel algorithm capable of detecting significant change points, or "points of interest" in an unsupervised fashion across multiple data streams in parallel. The algorithm is based on an incremental dimensionality reduction approach known as subspace tracking. Sensomax exploits this algorithm to detect the change points and dynamically respond to the applications' demands whilst executing concurrent applications, switching operational paradigms and reorganising at cluster and network levels.