Adaptive Anomaly Detection of Coupled Activity Sequences

In this paper, we employ agent technology to develop a pattern mining system to detect abnormal trading activity patterns in the three coupled sequences including buy orders, sell orders and trades. The system uses six Hidden Markov Model(HMM)-based models to model the trading activity sequences in different ways: three standard HMMs for modeling single sequences respectively; an integrated HMM model combining all individual sequence-oriented HMMs; a Coupled HMM reflecting coupled relationships among sequences; and an Adaptive Coupled HMM to automatically capture the significant changes of activity sequences. The above six HMMbased models compete with each other. The outputs generated by the best model are used as the final outputs of system. The rest of this paper is organized as follows. We present the system framework in Section II. After Section III introduces the modeling of trading activity sequences by HMMbased methods, Section IV provides the approach to identify abnormal activity patterns using six HMM-based models. The model selection and evaluation are introduced in Section V, and the experimental results are given in Section VI. Finally, Section VII concludes this paper.