Churn prediction on fixed broadband internet using combined feed-forward neural network and SMOTEBoost algorithm

Abstract

The case of churn becomes a critical problem that often happens in many telecom companies. In many real cases of churn, imbalance and outlier data usually occurs in the dataset. These problems in some cases, make the conventional data mining approach is less successful to create a churn prediction model. Therefore, it is a concern of telecommunication service providers to investigate and developed various methods to overcome the problems. This paper combines feed-forward neural network and SMOTEBoost Algorithm for churn prediction. While the former is used to overcome the problem of noise, the latter is used to overcome the problem of imbalanced class. The first technique performs data reduction of noise, and the second one performs the task of prediction. In addition, SMOTEBoost, which is a method that combines SMOTE and Boosting algorithm, well performs in classifying imbalanced class dataset without sacrificing the overall accuracy.