A nonlinear outlier detection method in sensor networks based on the coordinate mapping

Abstract

This paper designs a nonlinear outliers detection method based on the coordinate mapping. Because data in different coordinate systems have specific attributes, the data coordinates in different coordinate systems can be transformed by coordinate mapping. Then the stream data features in a sensor network can be extracted accurately by principal component analysis to improve the detection accuracy of abnormal data points. Clustering of convection data features is implemented to shorten the time of subsequent detection by rapidly classifying data. Finally, the difference point factor is used to detect the nonlinear outliers in the sensor network. Experimental results show that the maximum detection accuracy of this method can reach 97%, the maximum detection time required is only 15 s, and the maximum miss rate of this method is 1.32%, indicating that this method can effectively detect nonlinear anomaly points.