An adaptive spatial outlier detection algorithm with no parameter for WSN

Abstract

Spatial outlier detection in wireless sensor network (WSN) can detect the objects whose non-spatial attributes are significantly different from their spatial neighbors, so as to ensure the reliability and accuracy of sensor data before decision-making process. The main drawback of existing spatial outlier detection algorithms is high user-dependency, which is not suitable for dynamic WSN data. This paper proposes an adaptive spatial outlier detection algorithm with no parameter, which can calculate the number of spatial neighbors based on global stability, and also can find outlier detection threshold automatically. This paper also proposes a new measurement for the degree of outlierness: spatial local outlier value (SLOV), which reflects the significance of spatial outliers more intuitively. Experimental results on WSN data of Intel Berkeley laboratory show that the spatial neighborhood found by nearest neighbor algorithm based on global stability conforms to the actual spatial distribution, and the algorithm is proved to be viable. Also, from experimental results on WSN data of GreenOrbs, it can be observed that the proposed algorithm can determine the number of outliers automatically according to the data characteristics, and the detection accuracy is generally over 80%, higher than that of existing SLOM and SLOF algorithm. It can be concluded that the adaptive spatial outlier detection algorithm proposed in this paper can detect spatial outliers without any parameter specified and with comparatively high detection rate, so it is verified to be feasible for wireless sensor network.