A preliminary prediction model of untreated dental caries by using machine learning method: a population-based cross-sectional survey

Objectives This study aimed to establish a preliminary prediction model for untreated dental caries among adolescents by using machine learning method. Materials and Methods Data were obtained from a cross-sectional epidemiological survey involving 1,641 adolescents aged 12 years in Guangdong, Southern China. Demographic information, socioeconomic status, and caries-related behaviors were collected via a structured questionnaire. A preliminary prediction model was developed using six machine learning (ML) algorithms. The performance of each algorithm was evaluated through stratified 5-fold cross-validation, with the following metrics: area under the receiver operating characteristic (ROC) curve (AUC), accuracy, precision, specificity, sensitivity, and F1 score. Variable importance rankings were generated, and calibration and decision curves were plotted. Results Among six machine learning algorithms, random forest performed the best, which demonstrated an AUC, accuracy, precision, specificity, sensitivity, and F1 score of 0.785 (0.759–0.811), 0.707 (0.685–0.728), 0.698 (0.673–0.724), 0.714 (0.695–0.737), 0.699 (0.675–0.730), and 0.698 (0.666–0.730) in stratified 5-fold cross-validation. The top three important variables: self-evaluation of oral condition, have a toothache in the past year, and oral health knowledge. Conclusions The prediction model based on random forest algorithm could help discriminate adolescents with untreated dental caries, which will assist healthcare providers in the individual management of caries in adolescents.