Scholarlink

<p>This investigation introduces a hybrid framework combining the Walrus Optimization Algorithm with XGBoost to predict depression among students in interactive learning environments. The model demonstrates consistent performance with accuracy of 84.18%, precision of 84.13%, recall of 84.18%, and F1 score of 84.12%. Cohen’s Kappa reaches 0.6731, indicating substantial predictive reliability, while the R² value of 0.9504 confirms the model explains over 95% of variance in depression outcomes, with MSE of 0.1582 and RMSE of 0.3978. SHAP analysis identifies sleep quality as the strongest predictor with a contribution value of 0.21, followed by academic pressure at 0.18 and financial stress at 0.16. Critical thresholds emerge where sleep duration below 5&#xa0;h nightly corresponds to depression score increases of 1.2 points per hour lost. Students studying less than 1&#xa0;h or more than 10&#xa0;h daily exhibit 47% and 63% higher depression rates respectively. The optimization process converges within 75 iterations, reducing MSE from 0.156 to 0.1149 and identifying optimal parameters of learning rate 0.056, maximum depth 5, and 121 trees. Computational efficiency analysis reveals 127&#xa0;s training time and 0.8 millisecond prediction latency. This framework offers educational institutions a practical and interpretable tool for early mental health detection and targeted intervention strategies that ultimately enhance student’s quality of life.</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

Optimizing student depression prediction using WaOA-XGBoost: A bio-inspired approach

  • 期刊论文

  • 发表日期：

    2025年12月19日

查看全文
  • A. Tamilarasan

摘要

This investigation introduces a hybrid framework combining the Walrus Optimization Algorithm with XGBoost to predict depression among students in interactive learning environments. The model demonstrates consistent performance with accuracy of 84.18%, precision of 84.13%, recall of 84.18%, and F1 score of 84.12%. Cohen’s Kappa reaches 0.6731, indicating substantial predictive reliability, while the R² value of 0.9504 confirms the model explains over 95% of variance in depression outcomes, with MSE of 0.1582 and RMSE of 0.3978. SHAP analysis identifies sleep quality as the strongest predictor with a contribution value of 0.21, followed by academic pressure at 0.18 and financial stress at 0.16. Critical thresholds emerge where sleep duration below 5 h nightly corresponds to depression score increases of 1.2 points per hour lost. Students studying less than 1 h or more than 10 h daily exhibit 47% and 63% higher depression rates respectively. The optimization process converges within 75 iterations, reducing MSE from 0.156 to 0.1149 and identifying optimal parameters of learning rate 0.056, maximum depth 5, and 121 trees. Computational efficiency analysis reveals 127 s training time and 0.8 millisecond prediction latency. This framework offers educational institutions a practical and interpretable tool for early mental health detection and targeted intervention strategies that ultimately enhance student’s quality of life.