Research On Outlier Detection In Wireless Sensor Networks

In wireless sensor networks, no matter for event monitoring or for space-time data collection, we urgently need an online outlier detection method. However, traditional outlier detection techniques are not directly applicable to wireless sensor networks due to the nature of sensor data and specific requirements and limitations of the wireless sensor networks. In accordence with the spatio-temporal correlation of the data in wireless sensor networks, two outlier detection methods are proposed.Outlier detection method based on Naive Bayes classifier. First, based on the theory of super-ellipsoid geometry, a dataset similarity degree is given. Using dataset similarity degree to choose neighbor nodes is superior to using the distance between nodes. Secondly, availing of dataset similarity degree, we cluster the WSNs, and in each cluster a Bayes classifier is constructed. Our method improves the detection accuracy. At the same time, the error propagation can be limited in a certain range to avoid the destruction of the entire network. This model is a real-time online outlier detection method.Outlier detection method based on dynamic threshold. Using the temporal correlation an outlier factor is defined, and the dataset is divided into three parts by the outlier factor: normal state, critical state and abnormal state. There is a dynamic threshold in the outlier factor formula. On the basis of Markov assumptions, the type of the data and a large number of simulation experiments, the dynamic threshold updating mechanism is given. This model is an online outlier detection method.