Imbalance Classification Model for Churn Prediction

Churn prediction deals with challenging problem of detecting customers who probably cancel a subscription to a service. Data mining techniques such as decision tree, logistic regression, neural network are very successful in prediction customer churn. However, the prediction accuracy of these classification techniques reduces when handling with class-imbalanced data. Class-imbalanced data are common in the field of Churn prediction, mainly one or some of the classes have much more instances samples in comparison to the others. Classification techniques for imbalanced datasets usually correctly predict the results for the majority class, but do not perform well to predict results for the minority class. In this paper, we propose SMOTEBagging, which combines SMOTE sampling technique with Bagging algorithm to enhance the classification model to predict results for the minority class. The classification performance is obtained via 5-fold cross validation. The experimental results show the effectiveness of SMOTEBagging technique.