Prediction Method Based On The Loss Of The Hybrid Model Of Telecommunications Customers

Aims to Customer churn prediction problem of Telecommunications Industry. The paper is based on previous studies. The author focus on stable and accurate customer churn prediction model with four basic algorithms(C5.0, Neural Net, Logistic and SVM)from four aspects. Finally, an efficient customer churn prediction scheme is produced which are fit for provincial telecom operators.Use derived variables based on coding methods to improve effect of customer churn prediction model. In previous studies, prediction variable set is from customer database or calculated variables. The variables don't track relationship between customer consumption behavior change and customer churn tend. So the author introduces coding method to generate some of prediction variables. The empirical research results show that effect of prediction variable set containing with these derived variables is better than prediction variable set without containing derived variables.A new method for constructing customer churn prediction hybrid model is proposed. Because provincial telecom operators customers quantity is more than ten millions, various branch implement different marketing strategy, and brands of customers,account bills of customers and duration of customers are also different . Usually the same type of customer has similar consumption behavior. So it is necessary to construct customer churn prediction hybrid model based on different segmentation variables. The empirical research results show that prediction model of using branch as segmentation variable is best; and prediction model of using account bill as segmentation variable is secondary; and prediction model of using duration or brand as segmentation variable is worst.Voting Based on the Best Arithmeti(cVBBA)method is proposed to improve effect of customer churn prediction. In telecom customer churn prediction problem, customer churn rate is a very small proportion, usually that will pollute prediction model and lead to reduce prediction effect. The author uses VBBA method to solve the problem. The empirical research results show that the prediction effect of VBBA method is better than any single arithmetic model, so VBBA method is effectual to lift prediction precision of customer churn in telecom customer churn management.ROC curve is recommended to assess telecom customer churn prediction models. The author uses five assessment methods (accuracy, hit rate, gain graphics, lift graphics and ROC curve) to empirical research. The results show that ROC curve is the most excellent. It has three advantages: visual, integrated and exclusive.