Research About The Telemarketing Of Bank Based On Generalized Semiparametric Additive Models

Telemarketing appeared in America in the 1970 s firstly.Telemarketing of bank started from the late 1990 s,commercial bank established telephone banking centers one after another what can provide telephone banking service for customers,such as consultation,online service and complaint handling.Nowadays,commercial bank has received good results by active phone calls.Now we have entered the "Internet plus" era.Although the marketing mode is more abundant and convenient,advantages of telemarketing has always existed.Telephone marketing still has great potential for mining,such as strengthening the ability to identify target customers,improving the hit rate and reducing labor and time costs constantly.In order to get the maximum benefit at the lowest cost,it is a major concern of the major banks to establish an effective customer identification system.This article analyzes and models the actual retail banking data.The original data set comes from the telemarketing data of Portuguese retail bank in 2008 to 2013,including customer information,product information and the social and economic conditions.Before modeling,the data are descriptive figured out,the basic data characteristics and the prominent features of the potential customer information are mastered.Then,the selection of variables is carried out,and the semi-automatic feature selection is used in the feature selection,which is divided into two steps.The first step is the manual selection,and the second step is the automatic selection.The next step is to select models on the basis of knowledge of statistics and machine learning,comparing the random forest,neural network,support vector machine and generalized semi parametric additive models.The last step is to compare models.There are three evaluation indicators of the model,one is by drawing ROC curve to get the AUC value,the two is to calculate the LIFT value,and the three is to calculate the F-score value.It is found that the success of marketing customers is the feature attributes of potential customers is very obvious,For example,the customers are about 40 years old;the field of work is mainly concentrated in the administration,technology and bluecollar work;the degree of education tends to be higher education and so on.By comparing the forecasting results of four models,the prediction accuracy of the generalized semi parametric additive model is proven to be the highest.The evaluation indexes are superior to those of the other three models,so we can conclude that the optimal model to identify the potential customers is the generalized semi-parametric additive model.