Abstract: An important problem in anti-money laundering is to predict the possible transactions conducted by suspicious accounts. Markov model has a wide range of applications in economic predictions such as stock, commodity prices,market share and so on. But the prediction accuracy of the single markov model remains to be improved. A hybrid Markov model jointing with clustering,association rule and low order Markov model was proposed. In the process of constructing the model, the confidence-based pruning was conducted to reduce the time complexity. Finally, the model was used to predict the transactions among accounts in anti-money laundering. hhe experimental results show that this modcl has high prediction accuracy and is a good tradeoff between the prediction accuracy and the time complexity.

Key words: Hybrid Markov model, Prediction, Clustering, Association rule, Anti-money laundering