【Abstract】Background: Suicide is a severe global public health issue, with college students being particularly concerned due to their unique life transitions and psychological pressures, leading to higher rates of suicidal ideation and behavior. Objective: This study aims to construct a predictive model using machine learning methods combined with suicide psychological theories to improve the accuracy of identifying suicidal ideation among college students and analyze key risk and protective factors, providing a scientific basis for prevention and intervention. Methods: Using cluster sampling, data were collected from 13,889 college students at a university in Tianjin, resulting in 13,330 valid questionnaire. The questionnaire included 106 features such as socio-demographic characteristics and so on. Lasso regression was used for feature selection, and classification models were built based on decision trees, random forests, GBM, and XGBoost algorithms. Model performance was evaluated, and key factors were identified using feature importance functions and SHAP analysis. Finally, EFA was used to reduce dimensions and reveal underlying structures. Results: The XGBoost model performed the best, with an AUC of 0.838 and a sensitivity of 0.799, identifying 16 significant influencing factors. Through EFA, three core factors were discovered: negative experiences towards the self, negative feelings towards others, and protective factors. The findings support some of the opinions of the cubic model of suicide, escape theory of suicide, and the interpersonal theory of suicide, as well as making connections with Watts' concept of sense of connectedness. Conclusion: This study not only improved the accuracy of identifying suicidal ideation among college students but also provided new perspectives for crisis prevention and intervention in colleges. Additionally, the study explored a potential path for addressing the explainability issues of machine learning in the field of suicide research.

College Students,Suicidal Ideation,Machine Learning,Risk Factors,Protective Factors