Research On The Application Of Outlier Detection Model In Anti Money Laundering

Financial transaction network is a very complex graph network, to mining the small mode which is different from other data mode from the data which is massive,complex,containing noise data in transaction network, the same like the determination of suspected accounts in financial transaction network and research on anti_money, which can be transformed into a problem in data mining for outlier detection. The traditional outlier detection method based on clustering has two big shortcomings, on the one hand, differences for the class of data objects, there is no good model analysis, on the other hand, the link between classes, there is no suitable discovery method.In view of the above problems, we can combine the clustering, local outlier detection and link discovery together, and combine the relevant theory, and give a complete algorithm of outlier detection model.For each trading account, to make the regular initial data suitable for outlier mining,we can deal the data with the number of transactions and transaction amount,In order to make the regular data set more effective and complete attention should be paid to remove some invalid data.Using the two step clustering algorithm to cluster the normalized data set, will have similar characteristics of trading account into a class, the clustering algorithm has good clustering effect, also can find some abnormal data objects, the data set which is already clustered made the next analysis mode more targeted, and for each cluster were established for anomaly detection model based on Clustering: account differences on the within class, local outlier detection algorithm based on Mahalanobis distance calculation and outlier index presented in ascending way of transactions between the contact, take link discovery based on graph entropy reduction technique for clustering after trading networks the key nodes are given in the network transaction transaction. And through two groups of experiments of two step clustering accuracy improved clustering algorithm based on density and accuracy of outlier detection algorithm to verify the Mahalanobis distance.According to the project financial analysis system, in line with previous studies and implements a hybrid model for outlier detection module, verify the validity of the data according to the real case.