Development and evaluation of a multi-model stacking approach for caries risk assessment in adults using supervised machine learning

Background Dental caries is a chronic disease that requires intervention to prevent complications and minimise costs. Accurate caries risk assessment is crucial but traditional methods depend on skilled clinicians, limiting scalability. Machine learning offers an efficient and standardised alternative.Objective This study aimed to develop and evaluate computational models using machine learning techniques for predicting caries risk in adults.Methods A systematic review identified seven predictors that were applied to 3,000 balanced Universiti Teknologi MARA patient records spanning low, moderate and high caries risk. Seven algorithms (decision tree, XGBoost, k-nearest neighbors, logistic regression, multi-layer perceptron, random forest, and support vector machine) were tuned with K-fold cross-validation and stacking ensembles to enhance performance.Results The two-model stacking approach achieved the highest accuracy (95.17%) and ROC-AUC (receiver operating characteristic-area under the curve) (99.78%), followed by the three-model stacking with a strong accuracy of 93.63% and specificity (96.82%). Among single models, random forest and extreme gradient boosting stood out with an accuracy of 90.47% and 90.20%, respectively, demonstrating robust classification performance.Conclusions Multi-model machine learning approaches showed high accuracy in caries risk assessment, offering reliable predictions while reducing reliance on extensive clinical judgment. These methods provide standardised risk predictions and support the integration of machine learning into clinical workflows to enhance prevention strategies and patient care.