%0 Journal Article
%T Machine Learning-Based Prediction of Dental Caries Risk Using Behavioral, Clinical, and Socioeconomic Determinants
%A Nelly-Beatriz Quispe-Maquera
%A Betsy Quispe-Quispe
%A Kandy-Faviola Tuero-Chirinos
%A Lizbeth Acero-Condori
%A Claudia-Yessica Lauracio-Lope
%A Naysha-Inmaculada Paricoto-Chaiña
%A Daina-Katiuska Lopez-Quispe
%J Annals of Dental Specialty
%@ 2347-2022
%D 2026
%V 14
%N 1
%R 10.51847/Mj8jVNUgYd
%P 147-155
%X Dental caries remains one of the most common preventable oral diseases, yet many prevention programs still rely on broad age-based or service-based approaches rather than individualized risk estimation. This study aimed to develop and compare machine learning models for predicting 3-year incidence of new carious lesions using behavioral, clinical, and socioeconomic determinants. The central objective was to evaluate whether integrated multi-domain prediction improves identification of individuals who may benefit from intensified preventive care. A population-based oral health cohort with baseline and 3-year follow-up data was analyzed. Predictor variables included dietary sugar frequency, sugar-sweetened beverage intake, toothbrushing frequency, flossing habits, fluoride exposure, baseline DMFT, plaque index, salivary flow rate, household income, parental education, insurance type, and access-to-care indicators. Logistic regression with elastic net regularization, random forest, extreme gradient boosting, and a shallow neural network were trained using repeated cross-validation. Model performance was assessed using AUC, sensitivity, specificity, calibration slope, and calibration-in-the-large. Extreme gradient boosting achieved the highest overall discrimination, with an AUC of 0.84 and sensitivity of 82% at the prespecified high-risk threshold. Calibration was acceptable after probability recalibration, indicating that the model could support risk-stratified interpretation rather than only rank-order classification. SHAP analysis identified baseline caries experience, sugar-sweetened beverage consumption, household income, plaque index, and irregular dental attendance as the most influential predictors. The findings suggest that behavioral and socioeconomic variables add meaningful predictive value beyond clinical indicators alone. The best-performing model therefore provides an interpretable framework for identifying high-risk individuals before additional cavitated lesions develop. The study is limited by the use of a single regional cohort, potential measurement error in self-reported behaviors, and absence of external validation. However, the results demonstrate the feasibility of integrating clinical, behavioral, and social determinants into machine learning-based caries prediction. This approach may support preventive dental care, community screening, and public health planning when validated across more diverse populations.
%U https://annalsofdentalspecialty.net.in/article/machine-learning-based-prediction-of-dental-caries-risk-using-behavioral-clinical-and-socioeconomi-dex0lbvha4lplmx