The Assessment Of Bank Customer Money Laundering Risk Based On UIB Decision-Tree Algorithm

With the development of economic globalization and international financial architecture,money laundering is a grievous problem in global.Money Laundering refers to the behavior of legalizing the illegal income by the methods of trading or transferring to escape the regulation punishment and enjoy the fruit of crime.Money laundering has become an important aspect of endangering social stability and safety,because it not only damages the global finance but also connects the predicate and lower crimes.On one hand,it covers for the predicate crime;on the other hand,it provides economic support for the lower crime,which disrupts social stability.Hence,Anti-Money Laundering has become a new hot research field.Nowadays,China's anti-money laundering work is mainly based on the suspicious transactions data from grassroots financial institutions.But this suspicious transactions reporting system has some problems,such as fuzzy criterions,massive data,a high false alarm rate,easy to avoid and lack of adaptability,which leads to a lower reliability and validity of the reportedly suspicious transactions data.In this paper,a novel data mining algorithm is proposed according to the feature of bank customers and is applied to assess the money laundering risk of bank customers.This application can reduce the workload of the suspicious transactions recognition greatly,which has an important significance to perfect the suspicious transactions reporting system.Decision tree algorithm is a kind of predictive model which generates tree structure based on inductive algorithm.It has been applied in different fields such as bank credit evaluation,drilling engineering fault monitoring,and chemical production smoothness analysis.The dissertation aims at application of decision tree algorithm in anti-money laundering.A money-laundering risk rating model is explored based on decision tree algorithm to rate the money-laundering risk of accounts automatically.This model is to exclude accounts with low money laundering risk from huge amounts of accounts,which can make the financial institutions' job focusing on high risk accounts and promote the development of anti-money laundering.The traditional decision tree algorithm is unbiased when selects the decision attribute,so the generated rule set is not concise.To overcome the drawback,an Unbiased Iterative Brunching decision tree algorithm(UIB for short)is proposed.A corrected parameter is introduced in UIB to modify the information entropy of attributes.The modification ensures that the decision attribute has the greatest information entropy among all attributes.Moreover,the traditional decision tree algorithm cannot assess the money laundering risk of bank customers with missing attributes while the actual situation is that the data of bank customers is not always complete.To solve this problem,the UIB algorithm records the probabilities of different attributes in the process of constructing decision tree,and estimates the value of the missing attributes using statistical theory.The improvement can ensure the success assessment of the risks.After that,we use the UIB algorithm to design and realize the money-laundering risk rating model.Experiments are given to compare the performance of UIB and ID3 algorithms.The results show that UIB can evaluate the level of risk of the accounts which lack of property with a high accuracy.Another rule we can found from the experimental results is that the assessment accuracy is related to the size of the training set and the number of risk levels.The greater the size of the training set is,the higher the accuracy is.In addition,the accuracy is increasing with the increasing number of risk levels.