Study on Customer Churn Prediction Methods Based on Multiple Classifiers Combination

Abstract

Combining multiple classifiers combination, sampling techniques, and more appropriate evaluation metrics, we first compare the selection of multiple classifiers combination based on GMDH(S-GMDH) and other classification methods on nine class imbalance data sets; we analyze the change of classification performances with and without using sampling. Then we further do customer churn prediction on ‘churn’ from the nine data sets. It is concluded that class imbalance has severely affected classification performances of various classifiers, which will surely influence churn prediction. Experiments prove that it is an effective way to improve churn prediction by combining S-GMDH and sampling techniques.