The Research On The Application Of Knowledge Discovery In Databases In Anti-Money Laundering Field

Knowledge Discovery in Databases (KDD for short) is becoming a hotspot in Artificial Intelligence and Database domain. So far, research on KDD has covered several domains and techniques, such as time sequence rules, association rules, and classification rules. KDD combined with On-Line Analytical Processing and Data Warehouse is also new research branch. With the rapid development of Internet, Web KDD is also becoming a new hot research field.This dissertation focuses on the application of KDD in Anti-Money Laundering.Banks have accumulated a mass of transfer records since they used computer to deal, but those records are mainly used in simple query and statistic. Application of KDD in it is limited on Customer Relation Management, Financial Market Analysis, etc. Money laundering is a grievous problem that financiers and officers meet now. It's a kind of severe crime and appeared more and more rampant these years. Present techniques against money laundering are still out of date and they mostly depend on human to look for suspicious accounts by experience. There hasn't appeared any KDD method used in it.This dissertation aims at application of KDD in anti-money laundering. A mathematics model that fits anti-money laundering is explored. We also make great efforts on developing KDD methods to find the accounts that probably have laundered money. Research on automatic anti-money laundering is wished to progress. The followings are results of our research.First, we proposed an applying system frame that can fulfill common KDD and find money-laundering modes and doubtful accounts with financial domain knowledge.Second, aiming at the particularity of money laundering in finance domain, we prove that preprocessing data with domain knowledge can make data mining more effective. We also propose a set of statistical method to preprocess data in financial field. The experiments we did proved them right.Third, we propose an improved clustering analysis by using ants climbing hill. It can find the outliers, which are concerned.