Objective: To construct an interpretable machine learning prediction model for medical students short-video addiction and to identify the core influencing factors, in order to provide evidence-based basis for accurate screening and intervention of medical students short -video addiction behavior. Methods: In January 2025, a convenience sampling method was used to select 601 medical students in a medical college from Nanyang in Henan Province. Data on short-video addiction, psychological status and other related information were collected by using Short-form Video Addiction Scale, Depression Anxiety Stress Scales-21 (DASS-21), UCLA Loneliness Scale (3 edition), Self-control Scale (SCS), Perceived Social Support Scale, Medical Student Stress Scale (MSSS). Interpretable machine learning models were constructed based on K nearest neighbor (KNN), support vector machine (SVM), lightweight gradient boosting machine (LightGBM) and extreme gradient boosting (XGBoost), and the model interpretation was performed by SHapley Additive exPlanations (SHAP) value method. Results: Lasso regression identified four predictive factors: SCS score, DASS-21 depression/anxiety score, and MSSS score were four predictors. The comparison of model performance showed that LightGBM had the highest Area Under the Curve (AUC) (0.905), followed by SVM (0.899), XGBoost (0.892) and KNN (0.890). The Brier score of LightGBM was 0.120, and the calibration curve showed that the predicted probability was in good agreement with the actual probability. SHAP analysis showed that the importance of the predictors of short-video addiction of medical students was as follows:DASS-21 anxiety score, DASS-21 depression score, SCS score and MSSS score. Conclusions Both LightGBM and SVM models have good performance, demonstrating preliminary potential for identifying individuals with high risk of addiction. Depression, anxiety, and academic stress are the main positively correlated factors, while self-control ability is a negatively correlated factor.