Credit card companies calculate an accurate credit score by utilizing the personal information and credit data of new applicants. To analyze and predict credit ratings, there have been many studies using machine learning. However, previous research had limitations in improving prediction accuracy using single algorithms such as ensembles or deep learning and could not consider the problem of multiple histories of the same customer using different cards. This study proposes a hybrid algorithm that combines heterogeneous ensembles and TabNet, a deep learning algorithm specialized in tabular data, to address these issues. The study conducted comparative experiments with several state-of-the-art machine learning algorithms that have been used for credit card delinquency prediction.