Utilizing multi-level convolutional neural networks to achieve refined modeling and visual analysis of college students' mental health data.

Early identification of students' mental health issues has become an urgent priority in education and public health. However, existing studies often rely on questionnaire-based assessments or traditional machine learning models, which are limited by manual feature design and weak ability to capture the multidimensional and dynamic characteristics of psychological data. This creates a research gap in developing more adaptive and automated approaches for reliable prediction and monitoring. To address this limitation, the present study proposes the use of Convolutional Neural Network (CNN) for mental health modeling, taking advantage of its capability to automatically extract hierarchical features from multimodal inputs. For comparative purposes, Gradient Boosting Decision Tree (GBDT) and Support Vector Machine (SVM) are also implemented as baseline methods. A dataset combining academic performance, emotional fluctuations, social behavior, and lifestyle indicators was preprocessed and used for experiments.Results demonstrate that CNN achieves the highest predictive accuracy of 94%, compared to 89% for SVM and 87% for GBDT. Beyond accuracy, CNN also shows faster convergence and greater robustness across k-fold cross-validation. These findings highlight the significance of CNN as a more powerful tool for handling high-dimensional psychological data. The study contributes to bridging the gap between traditional mental health assessment and intelligent data-driven approaches, providing practical value for early risk detection and personalized interventions among students.