An anomaly detection approach based on symbolic similarity

Anomaly detection technology has two broad categories: segmentation based and kernel based anomaly detection techniques. According to different similarity measures,kernel based anomaly detection techniques including KNNC (k-nearest neighbor for continuous time series) and KNND (a discrete version of KNNC) method. KNNC has better accuracy but lower efficiency than KNND, but KNND would lost information in some cases. In this paper, we proposed a symbolic similarity based anomaly detection approach ANOKP which used a symbolic similarity KPDIST. KPDISP get a better accuracy than SAX through selecting Key Points in SAX discritizing result of time series. Experimental results on several real life data sets indicate that the proposed anomaly detection method ANOKP have better accuracy than KNND and similar efficiency with KNNC.