Design Of Bank Anti-money Laundering System Based On Anomaly Knowledge Discovery And Incremental Learning

Knowledge Discovery in Databases ( KDD) is the research focus of Artificial Intelligence and Data Mining in recent years, which covers a wide variety of knowledge discovery methods in many related domains and can discovery various of knowledge such as associated knowledge, classification knowledge, clustering knowledge and time series knowledge and other types of knowledge. In particular, along with extending and deepening of database application, discovery of knowledge from huge data bases is becoming more and more urgent, which further promotes the research and application of KDD.With the development of networking and informatization construction of bank, vast amounts of transaction data have been accumulated, while the use of data are still mainly focused on routine inquiries and statistical treatment, as regards the application of KDD, still at its initial stage, mainly focusing on customer relationship management and analysis, and market analysis. On the other hand, security and order of financial transactions faces increasing more and more threat, money laundering is a prominent one, which is a serious threat to security and order of financial transactions, as well as to national economic security, while anti-money laundering technique is still-backward, which is still in the stage of human intervention and urgently needs introduction of new related technologies to improve the efficiency of anti-money laundering. In respect of China's anti-money laundering work, the existing anti-money laundering system under the guidance of People's Bank of China mainly includes the work as follows: the banks and other related financial institutions report large and suspicious transactions, then relevant regulatory bodies finish related statistical analysis and subsequent processing of these reports, which has prominent defects: mass of reporting data, high rate of false positive, high rate of omission, easy circumvention of rules and low adaptability.In this paper, on the basis of practice and research of the existing anti-money laundering system, comprehensively utilizing related anomaly knowledge discovery techniques (Classification, Clustering and Rough sets learning theory etc.), we designed knowledge discovery algorithms suitable for anti-money laundering and then we design a more efficient bank anti-money laundering system with high adaptability using incremental learning. Our research is of great significance for promoting anti-money laundering research, improving China's current anti-money laundering systems, and efficiently carrying out anti-money laundering work.The main research done in this paper is as follows:Firstly, knowledge discovery and data mining, as well as the basic concepts, theory and state of the art of anti-money laundering were reviewed, mainly focused on the basic framework of anti-money laundering system and typical data mining algorithms used in Anti-Money Laundering field.Secondly, in view of mass, high sparse and uneven distribution of datasets in financial anti-money laundering field, based on rough set theory and existing research results of related decision tree classification algorithms, a novel algorithm called Decision Tree Algorithm Based on Decision Classify-Entropy was put forward to finish the task of mining rules in Anti-money laundering environment. On the while, according to the characteristics of clustering analysis and detection in Anti-money laundering field, on basis of existing research results of clustering algorithm and dissimilarity measure, a new money laundering detection and clustering analysis algorithm based on improved minimum spanning tree clustering was proposed, which could make us further focused on suspicious money laundering transactions and provide suspicious level of those transactions for further analysis.Thirdly, using the algorithms designed in this paper, around organization, building and incremental learning of knowledge base, we designed a new bank anti-money laundering system based on anomaly knowledge discovery and incremental learning and then developed a prototype of this system to test and evaluate the proposed algorithms and the bank anti-money laundering system designed in this paper.Finally, we summarized our research work, analyzed the contents which need intensive research and then looked forward to the research direction and perspective of Anti-money laundering.