A Machine Learning Application to Predict Customer Churn: A Case in Indonesian Telecommunication Company

Pp: 144-161 (18)

* (Excluding Mailing and Handling)

Abstract

This study aims to develop a churn prediction model which can assist telecommunication companies in predicting customers who are most likely subject to churn. The model is developed by employing machine learning techniques on big data platforms. Customer churn is one of the most critical issues, especially in high investment telecommunication companies. Accordingly, the companies are looking for ways to predict potential customers to churn and take necessary actions to reduce the churn. To accomplish the objective of the study, it first compares eight machine learning techniques, i.e., ridge classifier, gradient booster, adaptive boosting, bagging classifier, k-nearest neighbour (kNN), decision tree, logistic regression, and random forest. By using five evaluation performance metrics (i.e., accuracy, AUC score, precision score, recall score, and the F score), kNN is selected since it outperforms other techniques. Second, the selected technique is used to predict the likelihood of customers churning.

Keywords: Customer churn prediction, Churn, k-nearest neighbour, Machine learning, Telecommunication company.

Cite as