Machine Learning‐Based Prediction of Dental Caries in 12‐Year‐Old Adolescents: A Nationwide Cross‐Sectional Study in Korea
Gahyun Cho, Hae‐Young Kim, Ki‐Bong Choi, Sunmi Song, Junesun KimABSTRACT
Objective
We aimed to develop and evaluate machine learning models to support population‐level risk stratification for dental caries in the permanent dentition of 12‐year‐old Korean children and to visualise the relative importance of key predictors.
Methods
This cross‐sectional study analysed data from the 2021–2022 National Survey on Children's Oral Health, which included 18,671 children aged 12 years. Dental caries was assessed in accordance with the World Health Organisation Oral Health Surveys: Basic Methods. To predict the presence of dental caries, established caries risk factors were identified based on previous studies. Predictive models were constructed using regularised Logistic Regression (LR), Random Forest (RF), Support Vector Machine (SVM), XGBoost and CatBoost algorithms, and model performance was evaluated using the area under curve (AUC) and additional metrics. The importance of key predictors was visualised using SHAP values.
Results
The study included 18,671 participants aged 12 years, of whom 51.6% were boys and 48.4% were girls; 92.9% resided in urban areas. Overall, 6.9% of participants had dental caries in permanent teeth. Ridge LR achieved the highest performance on the weighted test set (AUC: 0.724, 95% CI: 0.690–0.754). The most important predictors of dental caries, ranked by importance, included the absence of sealants, poorer self‐perceived oral health, unmet dental treatment needs, lack of preventive dental care within the past year, girls, gingival bleeding, lack of oral health education, absence of orthodontic appliances, non‐use of dental floss, and tooth pain or discomfort. Furthermore, dental caries was more likely among individuals with lower perceived household economic status, higher consumption of beverages and snacks, residence in rural areas and a lower frequency of tooth brushing.
Conclusion
Findings underscore the importance of preventive oral behaviours; integrating predictive models into public health initiatives could help target interventions and improve oral health outcomes in children.