Prediction Of Customer Churn In Commercial Banks Based On

With the rapid development of our country’s economy, as an importanteconomic development auxiliary industry, the banking sector was rapid rise. At thesame time, more and more all kinds of levels banks emerging, which intensified thecompetition of banking. Intense competition led to the loss of bank’s customers, byresearch this issue, the author found that banking product homogeneity is serious, thecustomer ’s choice of products and services is growing, customers loyalty becomemore and more low. So predict the flow of customers, make timely and effectivecustomer management were the best way to reduce the loss of customers.In this paper, the author combines with working practice, and based on analyzedecision trees, neural networks, Bayesian and Support vector machine (SVM) theory,research the reason for customer churn, the research include research status andproblems, the analysis of the problems such as the status quo of the application,found a non-equilibrium nuclear SVM what is suitable for application to predictionsolve the problem of Bank of Xinxiang(local banks) customer churn. The studydefine a churn related variables according to the Bank staff survey and collectionbanks existing data. Using related principles to analysis the existing data bycontinuous variables and discrete variables. Noise processing was used in theanalysis，the analysis results were used in Bank of Xinxiang customer churnprediction problem. According to the experience of the bank staff and the datasampling of Bank of Xinxiang, define the related variables about customer churnproblem, analysis based on the sampling data of continuous variables and discretevariables, and then make the noise processing, Logistic regression with SAS9.2multilingual version programming operation, forecasts from the initial screening ofmore than20variables determined5the largest contribution to the predictionvariables, and then use the non-equilibrium nuclear vector machine (SVM) algorithm,get loss probability value. Using customer churn probability values to recordcustomers loss prediction scores. Put predict outcome score from big to small by theabove information, create customer churn scale table, compose the bank customer churn prediction result set. Selected some customers form the prediction result setwhich had the probability of a larger value as the final loss of customers, providedthe result to the relevant personnel of the bank, let them make doula care tocustomers and retention activities to minimize customer churn, make sure that thebank could be stable and development.