Clustering And Its Application On Evil Overdraw Forecast Of Credit Card

In information explosion's time, the mass data gushed out subsequently.This brings knowledge discovery with its own challenges, and provides a very big development space for data mining technology. Clustering is an important data mining technology, researched and applied widespreadly.Cluster can not only analysis the data independently, but also can combine with other methods to unearth more valuable knowledge.Clustering analysis originats many other scientific fields, including statistics, biology, data mining and machine learning, etc. At present, the clustering analysis method is mainly applied in market analysis, pattern recognition, data analysis, image processing and so on,which provides certain research subjects and corresponding technical support for the researchers and policymakers.With the development of the data mining for many years,the clustering has five methods.In practice, we can use a kind of method to solve the problem.But in mostly, we analyse the problem size and the data distribution, and combin the multiple clustering method thoughts, or introduce other aspects of data mining algorithm, and design a more efficient algorithm.The purpose is to find an efficient method to handle a variety of practical problems, and get more valuable knowledge.This paper analyzed two clustering algorithm commonly.First of all, it introduced the CLIQUE clustering algorithm, which is a cluster analysis algorithm based on the density and based on grid data.The algorithm has high efficiency for the large database in high-dimension, and can get very good clustering results. But we often simplify the method,so the accuracy of clustering results may be reduced. This paper analyzed the many characteristics of the CLIQUE algorithm, and put forward its shortcoming.This will point out the direction for the future research work. Then, the paper introduced mainly the K_means algorithms based on the division and its application. The algorithm utilizes heuristic method, and it has clear thinking to understandable, and it has fast convergence rate and the wide range of application presently.But the K_means algorithm also exists some shortcomings and the insufficiency:(1)the algorithm chosed the initial clustering center of the cluster analysis with random generation; (2) the algorithm is inputed more parameters. The selection of parameters effects the clustering results directly.Meanwhile it is a problem to choice the parameters.The paper studies the importance of the K_means algorithm clustering center initialization, as well as some existing initialize clustering center of algorithms and its improvement ideas. On this basis, we analysised two old methods based on distance and density of basic idea, and found a new method based on high density of initial clustering center algorithm. Experimental results show that based on high-density choose initial clustering center algorithm, can reduce K_means algorithm of iterations, and select more reasonable on the initial clustering center.Finally,this thesis has discussed the improvement k_means cluster analysis technology, and applied it to detect the evil overdraw of credit card. The model forecast the possibility which customers'intention to evil overdraw, and it can supply the forecast information to the bank, which also can take precautions against the defect phenomenon.