Outlier Detection For Multivariate Spatio-temporal Data In Wireless Sensor Networks

Data quality is very important for the application scenarios based on wireless sensor networks. Outlier detection controls the quality of measured data, improves robustness of the data analysis under the presence of noises and faulty sensors, and also provides an efficient way to search for values which can be treated as events indicating change of phenomenon that are of interest. However, traditional outlier detection techniques can not be directly applicable to wireless sensor networks due to the nature of sensor data and speci?c requirements and limitations of the wireless sensor networks.This paper provides a general overview and discusses several important aspects of existing outlier detection techniques speciflcally developed for the wireless sensor networks, and presents a technique-based taxonomy and comparation about them. After that, proposes an outlier detection algorithm called SlopeLOF for the multivariate spatio-temporal sensor data, which current approaches can not handle effectively. A degree, namely Slope Vector, is introduced to describe the variation tendency of multivariate spatial-temporal data for the first time; after a pre-processing method for raw sensor data depended on the temporal attributes and spatial attributes in wireless sensor networks, Mahalanobis distance is calculated for the similarity between Slope Vector of the pre-processed data; the algorithm obtains multivariate spatial-temporal outlier score by calculating classical LOF based on local density with Slope Vector. It is shows that the SlopeLOF algorithm can identify multivariate spatial-temporal anomaly data effectively in the test sensor data sets collected from our prototype experimental project.And then, the Receiver system design and implementation—a complete solution for real-time data collection, processing, storage, and online publishing for wireless sensor networks has been given. This system provides a rich set of data samples from our large-scale and long-term WSN application project, which are used in the SlopeLOF outlier detection. The general and flexible system architecture makes Receiver easy to adapt to more different application domains based on wireless sensor networks.