Outlier Detection of Data in Wireless Sensor Networks Using Kernel Density Estimation

In this paper, an attempt has been made to develop a statistical model for the sensor data stream, estimating density for distribution of data and flagging a particular value as an outlier in the best possible manner without compromising with the performance. A statistical modeling technique transforms the raw sensor readings into meaningful information which will yield effective output, hence offering a more reliable way to gain insight into the physical phenomena under observation. We have proposed a model that is based on the approximation of the sensor data distribution. Our approach takes into consideration various characteristics and features of streaming sensor data. We processed and evaluated our proposed scheme with a set of experiments with datasets which is taken from Intel Berkeley research lab. The experimental evaluation shows that our algorithm can achieve very high precision and recall rates for identifying outliers and demonstrate the effectiveness of the proposed approach.