Understanding the semantics of sensor data

Our system architecture to manage sensor data is described. Our data mining applications require past history of the sensor data. Therefore, unlike most present systems that focus on streaming data, and cache a small window of historic data, we store the entire historic data. Several interesting problems arise in these scenarios. We study two of them: (a) Given that a sensor can send data corresponding to its current configuration at any particular instant, how do we define the data that should be stored in the database? (b) Sensors try to minimize the amount of data transmitted. Also there could be data loss in the network. So the data stored will have lots of "holes". In this case, how can an application make sense of the stored data? In this paper, we describe our approach to solve these problems that enables an application to recreate the environment that generated the data as precisely as possible.