A Hybrid Classification Model Of Proportional Hazards Model And Back Propagation Artificial Neural Network

Recently,with the popularization of 5G,the promulgation of the number portability policy and the more homogenous products,the cost of customer churn becomes increasingly lower.For operators,the loss of the customer has a negative influence on their performance and reputation.Therefore,it is valuable to analyze the customer-related data collected by operators to find out the potential causes of customer churn and build a customer churn prediction model.In fact,the prediction of customer churn problem can be regarded as a classification problem.This thesis applies the survival analysis theory and machine learning method theory.A new hybird model is established based on the proportional hazard model and Back Propagation Artificial neural network.The key points of the new hybird model can be explained as follows:firstly,the collected data is processed with extreme value normalization method,and then the PH model coefficients are used for the normalized data transformation,next the Back Propagation Artificial neural network model is used for classification.Secondly,this thesis verifies the new model by simulating experimental data,and comparing the classification results with the single model.The experimental results show that the prediction accuracy,F1-score and AUC value of the new model are all improved to some extent in most cases.Thirdly,this thesis uses the customer data collected during 25 months to conduct empirical analysis.Through analysis,we identify several important factors that affect the customer churn.And then we compare the hybird model with the single model.Based on the survival time and nonlinearity of the data dicussed in the hybird model,we find that the prediction accuracy and performance of the hybrid model are better than those of single model.Finally,with the result of empirical analysis,this thesis puts forward some practical suggestions in terms of marketing and customer management.