Anomaly detection algorithm based on pattern density in time series

Abstract

Anomaly detection in a time series has attracted a lot of attentions in the last decade, and is still a hot topic in time series mining. In this paper, an anomaly detection algorithm based on pattern density is proposed. The proposed algorithm uses the anomaly factor to identify top k anomaly patterns. Firstly, a time series is represented based on its key points. Secondly, the represented time series is partitioned into patterns set. Thirdly, anomaly factor of each pattern is calculated, and anomaly factor is presented to measure the anomalous degree of a pattern by taking into account the characters of its neighbors. Finally, Top k anomaly patterns are identified. The effectiveness of the anomaly detection algorithm is demonstrated with standard and artificial time series, and the experimental results show that the algorithm can find out all anomaly patterns with different lengths. © 2013 Springer Science+Business Media New York.