Development of a Lightweight Convolutional Neural Network for Student Mental Health Assessment

Abstract

To address the limitations and subjective biases of self-report scales in student mental health assessments, this study developed an automated mental health assessment framework based on lightweight convolutional neural networks (LCNN). This system integrates multi-source data to achieve a more objective and efficient assessment model. The main innovations of this research are threefold. First, a lightweight network structure is used to reduce computational complexity. Second, multimodal data, including text content and behavioral logs, is integrated. Finally, a dynamic rating system with real-time monitoring capabilities is built. The experiments were conducted on an automatically constructed dataset containing 5000 samples. The results show that the model achieves a precision of 94.2%, a recall of 91.8%, and an F1 score of 92.9%, which is an improvement of about 10% over traditional methods. The system has completed preliminary application validation in a campus environment, demonstrating its practical value and providing new technical support for mental health monitoring.