GRADIENT BOOST TREE AND CATBOOST METHODS IN BANK CUSTOMER CHURN MODELING

• Main Author: Anna, Catherina
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/81956
• Institution: Institut Teknologi Bandung

With the advancement of innovation and technology, more banks are being established. The increasing number of new banks has led to intensified competition among banks to attract as many customers as possible. Banks that fail to implement effective marketing strategies to attract customers may face the risk of losing them. The condition where customers leave the bank and no longer use its services is known as churn. To address this issue, banks need to find ways to identify customers who are likely to churn so they can focus their strategies on these customers. Tree Ensemble Learning models can be used to predict bank customers who will churn based on the available data. Customers who are likely to churn often exhibit certain characteristic patterns that can be observed in their data. These patterns will be analyzed to create a model that can predict which customers will churn. In this thesis, the methods used are Gradient Boosting Tree (GBT) and CatBoost. The GBT and CatBoost methods are applied to customer data that has undergone and has not undergone over sampling and under sampling processes. These methods are expected to give good evaluation metrics so it can be used in the future to predict customers who will churn. Both methods have been proven to produce very good evaluation metrics. Although the modeling with CatBoost produces a higher ROC AUC, the modeling with GBT yields higher accuracy, precision, recall, and F1 scores.