Comparative Evaluation of Machine Learning Models for Credit Card Approval Prediction Using Class Imbalance Handling Methods in the PyCaret AutoML Framework

Credit card approval processes are often performed manually, leading to inefficiencies, higher operational costs, and inconsistent decision-making. This study comparatively evaluates several machine learning models under different class imbalance handling strategies to predict credit card approval using the PyCaret AutoML framework. Two experimental scenarios were conducted on a dataset of 25,128 credit card applications, namely PyCaret's built-in SMOTE-based imbalance handling and random undersampling. In the SMOTE-based scenario, the Random Forest model achieved high accuracy, but its recall for the minority class was only 36.11%. In contrast, random undersampling produced more balanced performance, with the Gradient Boosting model achieving 100% recall. These findings indicate that class imbalance handling plays a critical role in credit card approval prediction, and that random undersampling provides more reliable minority-class detection than PyCaret's built-in SMOTE approach for the evaluated dataset.