Sensor Data Stream Exploration for Monitoring Applications

This paper presents StreamXPlore, a system that enables users to explore historical stream data in order to determine what events to monitor in the future. At the heart of StreamXPlore is a new event modeling mechanism. StreamXPlore enables the specification, analysis, and mining of these new types of events. Event analysis enables event refinement using data-cube-style slice, dice, drilldown, and roll-up operations. Event data mining consists of event clustering for pattern identification. Preliminary experiments on a real stream of hydro-sensor data show that the proposed exploration mechanisms can indeed be helpful in identifying interesting patterns and are sufficiently fast for exploration of gigabyte-size stream repositories.