Customer Churn Prediction using Machine Learning with Survival Effect Approach: An Empirical Study on Telecommunication Customer Behavior
Soni Adiyatma (a*), Heri Kuswanto (a), Dedy Dwi Prastyo (a), Endah Setyowati (b)

(a*) Department of Statistics, Institut Teknologi Sepuluh Nopember Surabaya
Jalan Raya ITS, Sukolilo, Surabaya 60111, Indonesia
*adiyatma.soni[at]gmail.com
(b) Faculty of Islamic Economics and Business, Institut Agama Islam Negeri Ponorogo
Jalan Puspita Jaya, Ponorogo 63492, Indonesia

Abstract

This research aims to obtain a predictive model of customer churn communication using a machine learning approach with a survival effect. Customer churn prediction is a process in business decisions by identifying customer behavior into the classes of churners and loyal customers. The methods used are machine learning (SVM and Random Forest), survival analysis (Cox Proportional Hazard), and machine learning with survival effect (Survival SVM and Random Survival Forest). The data used is the telecommunication customer behavior data at telco company X in 2019. The predictor variables used were gender, monthly charges, phone services, TV streaming services, senior citizens, dependents, and marital status. The result of data exploration showed that gender and phone services have no significant effect. The comparison of customer churn prediction methods showed that the performance of survival analysis models and machine learning with survival effects has a higher concordance index than machine learning methods. It showed that the higher concordance index in machine learning with survival analysis indicated that the model has an almost perfect discriminatory power. In addition, the combination of machine learning and survival analysis offers predictions of customer churn by considering the time the customer will churn.

Keywords: Churn Prediction, Machine Learning, Survival Analysis

Topic: MATHEMATICS AND STATISTICS