Prediction of periodontal disease in adults using tree-based machine learning models

Abstract

This study aimed to evaluate the predictive performance of tree-based machine learning models for periodontal disease in adults using data from the 7th Korea National Health and Nutrition Examination Survey (2016–2018). Data from 8,064 adults aged ≥40 years who completed oral examinations were analyzed. Periodontal disease was defined as the presence of periodontal pockets measuring ≥4 mm, and demographic, behavioral, and systemic health factors were included as predictors. Four models—decision tree, random forest, XGBoost, and LightGBM—were developed and compared based on accuracy, precision, recall, F1 score, and area under the curve (AUC). The decision tree model demonstrated the lowest performance (AUC=0.553), whereas LightGBM achieved the highest (AUC=0.658), followed by the random forest (0.626) and XGBoost (0.624) models. Feature importance analysis revealed that sex, age, education, income, obesity, and diabetes were the most influential determinants across models. Although the overall performance was moderate compared with that reported in previous international studies, tree-based machine learning demonstrated the potential for predicting periodontal disease at the population level, underscoring its applicability in preventive and clinical oral health research.