The Analysis And Prediction Of The Successful Road Of Bank Telemarketing

In the telecommunications industry developed today,the phenomenon of telephone marketing has already appeared in the streets,but people's acceptance of the telephone marketing is getting lower,marketing results tend to make the marketing staff exhausted?The results of this paper have important theoretical value and practical significance for the customer management of commercial banks,discovering valuable customers and maintaining customer loyalty.With the large data projection,the use of data mining technology to carry out precise marketing of more and more fields,this paper proposes the use of data mining methods,to predict the result of long term deposits in the telemarketing bank,the paper collects 41188 foreign bank telephone marketing data,analyzes 150 characteristic variables related to bank customers,products and socio-economic attributes,and then reduces to 21 variables by artificial semi-automatic selection.Because the resulting dataset is non-balanced,only 11.3%data is a successful record of telemarketing,in order to clarify the impact of the unbalanced dataset on the model,after the missing value preprocessing,the new equilibrium dataset was generated by Chawla SMOTE algorithm,compared with the effect of the balanced data set and the unbalanced DataSet training model,the results of the model predicted by the unbalanced dataset were found more biased towards the majority of the sample.Therefore,this paper uses a balanced dataset to model training and evaluation.This paper considers three classification models:Logistic regression model,decision tree and support vector machine,and the effect of classification is measured by using the value of precision and the AUC of ROC curves.The interpretation of Logistic regression taxonomy and decision tree fitting model is easy to understand,and the new data also have better predictions,whereas the support vector machine model is comparatively complicated,but it has better learning ability for linear and nonlinear problems.Because of this complexity,support vector machines are often able to provide precise predictions,and after training comparisons to determine the parameters or structure of each model,the accuracy of three models measured by the test set data is 47.3%,respectively,The values of AUC in 73.1%and 52.6%,ROC curves were 0.921,0.985 and 0.938 respectively.In the field of marketing,managers are more eager to identify customers with higher value,try to avoid wasting resources on some low-valued customers,so as to improve the input-output ratio,then hope that the predicted results are more accurate,and the AUC value of this article is not very different,according to the highest precision principle,select decision tree C5.0 classification algorithm to predict.