Yuni Yamasari, Abdul Khahar, Ricky Eka Putra, Made Suartana, Paramitha Nerisafitra, Anita Qoiriah
Credit card default risk prediction is becoming increasingly important in the financial industry to minimize losses and improve risk management. This study proposes the development of a credit card default risk prediction model by implementing hyperparameter tuning using Optuna and applying ensemble learning to boosting algorithms such as XGBoost, LightGBM, and CatBoost. The research stages begin with data preprocessing, which includes removing irrelevant features, handling missing data, outlier processing, encoding categorical features, creating new features, splitting data, balancing classes, and selecting features. The hyperparameter tuning process is carried out using manual search and Optuna, followed by the application of ensemble learning using stacking and voting methods to improve model performance. The results show that the ensemble model with voting using a combination of XGBoost, CatBoost, and LightGBM optimized with Optuna gives the best results with an AUC of 0.9763. After retraining with new data, model performance increased significantly, with an AUC reaching 0.9933. The resulting model is able to predict credit card default risk with high accuracy and can be used for continuous prediction of new data. © 2025 IEEE.
Universitas Negeri Surabaya, Department of Informatics, Surabaya, Indonesia
