Scholarlink

<p>The increasing demand for university mental-health services, coupled with limited professional resources, delayed risk identification, and the inability of traditional assessment systems to deliver timely and personalized interventions, creates a critical gap in effective psychological crisis management for students. This research proposes an integrated Intelligent Feedforward tuned–Deep Deterministic Policy Gradient (IFF-DDPG) model for student psychological crisis prediction and intervention optimization. The Psychological Crisis Risk Dataset (N = 5800, 17 features) is preprocessed through missing value validation, Z-score normalization for feature standardization, and Principal Component Analysis (PCA) for dimensionality reduction and multicollinearity mitigation. A Feedforward Neural Network (FNN) performs baseline supervised crisis risk estimation. An Intelligent Feedforward (IFF) Neural Network dynamically regulates learning parameters using reward variation, temporal difference error, actor–critic loss, and Q-value variance to stabilize training. The Deep Deterministic Policy Gradient (DDPG) agent learns a continuous intervention policy within a 24-dimensional state and 4-dimensional action space, forming a closed-loop adaptive decision mechanism. Statistical validation using independent t-tests confirms significant post-intervention reductions in stress, anxiety, depression, and total psychological distress indicators. The proposed model achieves 97.68% classification accuracy with strong recall performance, demonstrating reliable crisis discrimination capability. Implemented in Python using DL and reinforcement learning libraries, the model ensures computational reproducibility and convergence stability. The results confirm that integrating supervised risk modeling with adaptive reinforcement learning provides a robust, scalable, and effective solution for intelligent psychological crisis prediction and personalized intervention optimization in academic environments.</p> Graphical Abstract <p>Graphical abstract of the integrated psychological crisis prediction and adaptive intervention model</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

Reinforcement learning and neural network-driven student psychological crisis intervention strategy

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年6月18日

查看全文
  • Hao Zhou

摘要

The increasing demand for university mental-health services, coupled with limited professional resources, delayed risk identification, and the inability of traditional assessment systems to deliver timely and personalized interventions, creates a critical gap in effective psychological crisis management for students. This research proposes an integrated Intelligent Feedforward tuned–Deep Deterministic Policy Gradient (IFF-DDPG) model for student psychological crisis prediction and intervention optimization. The Psychological Crisis Risk Dataset (N = 5800, 17 features) is preprocessed through missing value validation, Z-score normalization for feature standardization, and Principal Component Analysis (PCA) for dimensionality reduction and multicollinearity mitigation. A Feedforward Neural Network (FNN) performs baseline supervised crisis risk estimation. An Intelligent Feedforward (IFF) Neural Network dynamically regulates learning parameters using reward variation, temporal difference error, actor–critic loss, and Q-value variance to stabilize training. The Deep Deterministic Policy Gradient (DDPG) agent learns a continuous intervention policy within a 24-dimensional state and 4-dimensional action space, forming a closed-loop adaptive decision mechanism. Statistical validation using independent t-tests confirms significant post-intervention reductions in stress, anxiety, depression, and total psychological distress indicators. The proposed model achieves 97.68% classification accuracy with strong recall performance, demonstrating reliable crisis discrimination capability. Implemented in Python using DL and reinforcement learning libraries, the model ensures computational reproducibility and convergence stability. The results confirm that integrating supervised risk modeling with adaptive reinforcement learning provides a robust, scalable, and effective solution for intelligent psychological crisis prediction and personalized intervention optimization in academic environments.

Graphical Abstract

Graphical abstract of the integrated psychological crisis prediction and adaptive intervention model