Statistical Modeling of Sensor Data and its Application to Outlier Detection

Abstract. Various applications rely on a continuous processing of data streams originating from a network of interconnected and collaborated sensors. The processing of those streams has turned out to be a difficult task as sensors only have limited resources and the data they produce is inherently uncertain and unreliable. In order to bridge the gap from raw, uncertain sensor readings to a meaningful model of the physical phenomenon observed, statistical modeling techniques have proved to be an adequate approach. By means of a statistical model, a wide range of sensor network related topics can be covered. In this work, we present an initial approach to tackle an important problem in sensor processing, namely the detection of outliers, with a statistical model.