Research And Application Of Clustering In Commercial Bank Customer Segmentation

Data Mining is a emerging decision-making support process,and clustering analysis is the core technology and very active research direction of the Data Mining.The customer segmentation is a component of the Customer Relations Management(CRM), which means that the enterprises classify the customers into different groups based on their characteristic,demand,fancy and other synthesis factors so as to provide the specific products and services to dissimilar customers.As the powerful tool in customer segmentation,the clustering methods are displaying its function of guidance in this field, especially in the financial industry.Clustering is a unsupervised learning process,it divides the data points into several parts by making objects in the same part have a high similar feasure and objects in different part are as dissimilar as possible.Clustering ensemble method are more robust and higher accurate by combining multiple clustering results.In this paper,we proposed a new two-layer clustering ensemble algorithm based on the voting mechanism to further improve the results of ensemble.This alogrithm mainly solve the following problem:Generation of the clustering members:the clustering members are the base of the clustering ensemble.The proper difference between members will get a better ensemble result.The model proposed involve two levels of cluster members:the members of the first level is generated by different original clusterig method with diverse parameters. Members of the second level are the clutering ensemble results of members of the first lever.It can improve the result of final data partition by select different kind of the original clustering algorithm and set diverse parameters.Clustering fusion function designation:the clustering fusion function is the method to combining the original clustering results.In our algorithm,a majority voting rule is adopted as the fusion function.The idea behind majority voting is that the judgement of a group members is superior to those individuals.This concept had been wdely explored and showed that it is a very simple,effective and easily be understood.According to the demand in real application,it put forward higher requirements for the model understandable.So we last chose the voting mechanism as the fusion function of each level.Matching the clustering label:it is a very important problem when adopting the voting-based clustering ensemble method,as different algorithms,even different parameters to a same algorithm will make totally different description to clusters which in fact are the same group.In this paper,this problem was solved by making the clusters which get the most shared data points have the same cluater label.This paper also discussed the universal method of customer segmentation.We also use the empirical analysis in customer segmentation.Based on the original transaction data of investment customers,we designed a clustering-classifer data mining process and built a effective customer segmentation model for the commercial bank.The result showed that the proposed two-layer clustering ensemble algorithm works very well in the business application.