Scholarlink

<p>This study investigates the key risk and protective factors influencing the mental health of young university students, aimed at providing empirical evidence for colleges to effectively identify high-risk students and formulate precise intervention strategies. Traditional psychological scales have limitations in prediction accuracy. Therefore, this study adopted qualitative methods to identify 14 characteristic factors reflecting the mental health status of university students. Multiple machine learning algorithms were then used to build prediction models. Through performance comparison, the Gradient Boosting algorithm was finally selected and combined with SHAP for feature importance ranking and influence analysis, to reveal the complex relationships between various factors and mental health risk. The ensemble learning models demonstrated superior performance. Specifically, the Gradient Boosting model achieved the highest overall performance with an accuracy of 0.923, an AUC of 0.877, and excellent cross-validation results. In contrast, Logistic Regression yielded an accuracy of only 0.707 and an AUC of 0.647. Key factors influencing mental health risk were psychological attention (SHAP value: 1.0839, a positive predictor), only child status (SHAP value: 0.2633, a positive predictor), and interpersonal distress (SHAP value: 0.1370, found to be a protective factor). Mental health risk in university students is influenced by multiple factors with varying directions of action. Accordingly, corresponding policy recommendations are proposed. This study hopes to provide a new paradigm for mental health assessment using machine learning and offer precise targets for mental health intervention in universities.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Predicting and Analyzing Mental Health Risk and Characteristics of Young University Students Using Machine Learning and SHAP

  • 期刊论文

  • 发表日期：

    2026年2月23日

查看全文
  • Xie Congying,
  • Wang Jiacan

摘要

This study investigates the key risk and protective factors influencing the mental health of young university students, aimed at providing empirical evidence for colleges to effectively identify high-risk students and formulate precise intervention strategies. Traditional psychological scales have limitations in prediction accuracy. Therefore, this study adopted qualitative methods to identify 14 characteristic factors reflecting the mental health status of university students. Multiple machine learning algorithms were then used to build prediction models. Through performance comparison, the Gradient Boosting algorithm was finally selected and combined with SHAP for feature importance ranking and influence analysis, to reveal the complex relationships between various factors and mental health risk. The ensemble learning models demonstrated superior performance. Specifically, the Gradient Boosting model achieved the highest overall performance with an accuracy of 0.923, an AUC of 0.877, and excellent cross-validation results. In contrast, Logistic Regression yielded an accuracy of only 0.707 and an AUC of 0.647. Key factors influencing mental health risk were psychological attention (SHAP value: 1.0839, a positive predictor), only child status (SHAP value: 0.2633, a positive predictor), and interpersonal distress (SHAP value: 0.1370, found to be a protective factor). Mental health risk in university students is influenced by multiple factors with varying directions of action. Accordingly, corresponding policy recommendations are proposed. This study hopes to provide a new paradigm for mental health assessment using machine learning and offer precise targets for mental health intervention in universities.