Outlier Detection Method Of Environmental Sensor Data Streams Based On Kernel Density Estimation

With the rapid development of wireless sensor network (WSN) technology, WSN has been applied to many fields. Environmental monitoring is a typical application of WSN. A great number of nodes have been deployed around the monitored environment, and collect the parameters (e.g, temperature, humidity, light, atmospheric pressure) so that the administrators can obtain the changes of environmental situation. In order to monitor all possible events (such as air pollution, landslides, forest fires), the sensor network need to analyze and process the multivariate data. Espeacilly, the network should be able to detect the abnormal data and issue the precausion signals real-timely. So the research on outlier detection in wireless sensor networks is of great importance for environmental monitroing.The main contribution of this paper is given as follows.1. An outlier detection method of environmental sensor data streams based on multimensional Epanechnikov kernel density estimation is proposed. The method consists of a sampling algorithm based on k-th nearest distance (ODBKND) and a kernel density estimation algorithm based on Epanechnikov (ODBKDE).2. The proposed outlier detection method has three steps. Firstly, the nodes remove all the abnormal data in nodes’ sliding window by ODBKND, and transmit the normal data to the cluster nodes. Secondly, the cluster nodes establish data distribution model by ODBKDE. Finally, the cluster nodes distribute the established model to the other nodes within the same cluster, and detect the outliers based on this model.3. The performance indexes are designed for evaluate the proposed algorithms. For the synthetic data, a series of experiments were carried out under MATLAB simulation environment, and the outlier detection algorithm based on histogram was compared with the presented algorithm. The simulation experimental results show that the detection accuracy rate of our algorithm is98%.