Performance Analysis of Oversampling and Undersampling on Telco Churn Data Using Naive Bayes, SVM And Random Forest Methods

Abstract

The telecommunications industry is a rapidly growing economic sector in today’s digital era. However, intense competition causes challenges in retaining customers. One of the problems faced by companies is the increase in customer switching, known as customer churn, which is when customers stop using the company’s services. To solve this problem, telecommunication companies use data mining techniques to predict potential customer churn. In the churn prediction process, data mining identifies patterns of customer behavior that indicate potential churn in the future. The dataset used in this research comes from the Telco Churn dataset which can be accessed through the Kaggle platform. This research implements oversampling and undersampling techniques to improve prediction accuracy on minority data in data mining algorithm models such as Random Forest, Naive Bayes, and SVM. The results of this study show that the application of oversampling and undersampling techniques is effective in increasing the Specificity (True Negative Rate) value of the algorithm model used. This Specificity increase is observed in various algorithms such as Random Forest which experienced an increase in Specificity value up to 45.98%, Naive Bayes up to 14.97%, and SVM up to 65.41%.