Abstract: Objective: To construct a risk prediction model for root caries in the elderly using machine learning, identify risk factors, and provide a basis for clinical screening and prevention. Methods: 4,826 patients aged ≥60 from Shanghai Yangpu District Central Hospital (2024-2025) were included. Data were collected via oral examinations, clinical records, and questionnaires. Risk factors were screened by multivariate Logistic regression. The dataset was split into training (70%) and validation (30%) sets. Four machine learning models were built using Python and evaluated effectiveness based on the area under curve (AUC). Results: The prevalence of root caries was 45.3%. Multivariate analysis showed 7 factors (e.g., age, diabetes) were significantly associated (P<0.05). The four models had AUC 0.849-0.894, with Logistic regression optimal and plaque index contributed most. Conclusion: Logistic regression had good predictive performance. Intervening on key risk factors can reduce root caries incidence in the elderly and improve oral health management.

Key words: root caries in the elderly, machine learning, Logistic regression, risk factors, prediction model