The Comparison of Random Forest and Artificial Neural Network for Customer Churn Prediction in Telecommunication

Customer churn is a significant issue that can affect a company’s revenue, particularly in the telecommunication sector. Companies today are working hard to survive in this competitive market by implementing various types of strategies. The churn prediction system can be used as a strategy for long-term business success or failure. In this research, variable reduction was carried out using the chi-square test and RFE to select several influential features to be used as input in the model. The implementation of the Synthetic Minority Over-sampling Technique for Nominal and Continuous (SMOTE-NC) is also carried out to handle the imbalanced data. This research compares the Random Forest and Artificial Neural Network algorithms to obtain the best accuracy in predicting telecommunication customer churn. In addition, Grid Search is used to perform parameter tuning to obtain the best parameters from each method. Therefore, this research obtained 4 models that were compared. From this experiment, it can be inferred that the Random Forest model with some hyperparameter tuning works best with an AUC (Area Under Curve) value of 75% compared to the Artificial Neural Network with an AUC value of 65%.