Outlier Geometric Angle Detection Algorithm

Abstract

Massive logs are generated in telecommunication networks. It is a challenge to analyze abnormal information in the big data logs quickly and effectively. We present a new outlier detection algorithm based on Unsupervised Learning Algorithm by geometric angle scanning judgment. First, calculate geometric center of measured data and several observation points around the measured data. Outliers can be segregated from normal area by density contrast method by angle based calculation. Results show that outlier geometric angle detection (OGAD) algorithm can separate anomaly from measured data effectively, and improve the accuracy of anomaly identification.