Objectives: To evaluate machine learning models used for predicting early childhood caries and identify key risk factors contributing to model performance. Methods: Systematic searches were conducted across major databases up to July 2025. Data extraction and risk of bias assessment were performed using PROBAST+AI tool. Model types, datasets, performance metrics, and validation strategies were analyzed. Results: Twenty-eight studies met inclusion criteria. Random forest models were most frequently used. Model accuracy ranged from 70% to 95%. Dental plaque and active caries lesions were the most consistent predictors. External validation was limited across studies. Conclusions: Machine learning shows strong potential for ECC risk prediction; however, standardized reporting and external validation are essential for clinical translation.
Atif et al. (Sun,) studied this question.