Telecom Customer Churn Prediction Based On Ensemble Model

The data mining technology(DMT)is approved by more and more people in the era of big data.It has become a trend that find out the underling rule behind data in the DMT.In the telecommunications industry,with focus of DMT,the prediction of telecom customer churn is one of a bit hot.Since current telecom customer churn prediction models are mainly based on traditional single models which face numerous function choices or parameter selections.Due to the lack of prior knowledge,the best configuration of the model is difficult to find.Therefore,the predictive performance of the model is not easy to improve.In previous studies,it was found that the ensemble model not only performs well in predicting results,but also avoids the problem of optimal configuration of parameters in a single model.So it provides a new idea for improving the forecasting performance of the model.In addition,to solve the high dimensionality problem in the telecom customer chum dataset,we have studied and compared the effect of different feature selection methods on model prediction performance,which is of great significance for improving modeling efficiency and predictive performance.In the research background of telecom customer chum prediction,an efficient and accurate chum prediction model will be established.Firstly,clearing the purpose of research.secondly,extracting the relevant user data from the database,which including user information of properties,call data,flow data,short message data and consumer data.Thirdly,with the step of data understanding,data preprocessing,data exploration and modeling evaluation,four churn prediction models based on decision tree,logistic regression,Bagging-CART and Adaboost-CART are respectively established.Finally,the efficiency and precision of these models have been analyzed and contrasted.The results show that:(1)The forecasting performance of the model was significantly affected by the different feature selection methods;(2)The ensemble model can significantly improve the forecasting performance compared to the single model;(3)Compared to the results obtained by no feature selection,the ensemble model based on information gain rate can significantly improve the generalization performance.