Handling imbalanced data in churn prediction using RUSBoost and feature selection (Case study: PT. Telekomunikasi Indonesia regional 7)

Abstract

Solving imbalance problems is a challenging tasks in data mining and machine learning. Most classifiers are biased towards the majority class examples when learning from highly imbalanced data. In practice, churn prediction is considered as one of data mining application that reflects imbalance problems. This study investigates how to handle class imbalance in churn prediction using RUSBoost, a combination of random under-sampling and boosting algorithm, which is combined with feature selection for better performance result. The datasets used are broadband internet data collected from a telecommunication industry in Indonesia. The study firstly select the important features using Information Gain, and then building churn prediction model using RUSBoost with C4.5 as the weak learner. The result shows that feature selection and RUSBoost improve 16% of the performance of prediction and reduce 48% of the processing time.