Research And Application Of Customers Churn Prediction Model

Along with the global business more competitive,customer churn prediction has become the most important content in customer relationship management.Predicting customer retention and formulating corresponding customer churn strategies has become a key factor in promoting enterprise development.Based on the analysis of customer behavior data,aiming at the redundant features and unbalanced positive and negative samples in customer data,this paper puts forward a new feature selection algorithm and unbalanced data processing algorithm,respectively,to establish a new customer churn prediction model.The specific contents of this paper are as follows:1.Aim to solving the problem of massive data and redundant features,this paper proposes a mRMR feature selection algorithm based on multi index fusion(MIF-mRMR).MIF-mRMR improves the mRMR algorithm and combine the Mahalanobis distance and maximum information coefficients to evaluate correlation between features and categories,and features.The experimental results show that the method has selected the feature subset dimension to be smaller and the accuracy rate is higher 3%than mRMR method.2.This paper proposes an imbalanced data classification method based boundary mixed sampling to solve the problem of the percentages of the churned customers differ considerably from non-churned.This method first introduces variable coefficient to find the boundary and non-boundary samples,and then deal with the minority samples in boundary for over sampling and majority samples in non-boundary for under sampling,separately,to achieve the basic balance of samples.Experiments results show that the proposed method owns the better classification performance than other three popular method.3.Based on the above method,this paper proposes a new customer churn prediction model.This model first use MIF-mRMR and BMS algorithm for feature selection and data equalization,respectively,and then put the equalized data into SVM,C4.5 and random forest classifier for customer churn prediction.The experimental results show that this model can get better customer churn prediction results,and meanwhile SVM is more suitable than other two classifier for the research of customer churn prediction.