Detection and Exploration of Outlier Regions in Sensor Data Streams

Sensor networks play an important role in applications concerned with environmental monitoring, disaster management, and policy making. Effective and flexible techniques are needed to explore unusual environmental phenomena in sensor readings that are continuously streamed to applications. In this paper, we propose a framework that allows to detect outlier sensors and to efficiently construct outlier regions from respective outlier sensors. For this, we utilize the concept of degree-based outliers. Compared to the traditional binary outlier models (outlier versus non-outlier), this concept allows for a more fine-grained, context sensitive analysis of anomalous sensor readings and in particular the construction of heterogeneous outlier regions. The latter suitably reflect the heterogeneity among outlier sensors and sensor readings that determine the spatial extent of outlier regions. Such regions furthermore allow for useful data exploration tasks. We demonstrate the effectiveness and utility of our approach using real world and synthetic sensor data streams.