Comparative analysis of CatBoost and LightGBM algorithms for predicting dental caries risk based on lifestyle factors

Chayadi Oktomy Noto Susanto (Universitas Muhammadiyah Yogyakarta)
Muhajirah Ulfah (Universitas Muhammadiyah Yogyakarta)
Bimo Aditya Pangestu (Universitas Muhammadiyah Yogyakarta)
Slamet Riyadi (Universitas Muhammadiyah Yogyakarta)

Jurnal Ilmiah Kursor
Vol. 13 No. 3 (2026)

Dental caries is a common oral infection but significantly impacts life. This study aimsto compare the performance of the CatBoost and LightGBM algorithms in predictingdental caries risk based on health and lifestyle data from over 55,000 South KoreanNHIS participants. Imbalanced data was addressed using the SMOTE technique, whilehyperparameter tuning was performed using Optuna to maximize the F1-score. Themodels were evaluated using the confusion matrix, accuracy, logloss, ROC Curve,AUC, Precision, Recall, and F1-Score metrics. The results showed that CatBoost wasslightly superior with an accuracy of 90.4%, an F1-score of 89.7%, and an AUC of 0.95,compared to LightGBM with an accuracy of 90.1%, an F1-score of 89.2%, and an AUCof 0.94. In addition, CatBoost also produced fewer false negatives, making it moresensitive in detecting caries cases. Therefore, CatBoost is recommended as the primarymodel in a caries risk prediction system to support early detection and diseaseprevention more effectively.