Enhancing Mental Health Detection Model using Adaptive-Based CNN

Abstract

This study proposes a Gaussian-CNN for mental health classification using questionnaire data collected from 150 university students. This study implements and compares four models, namely Logistic Regression, Support Vector Machine (SVM), CNN, and the proposed Gaussian-CNN. We find that traditional machine learning models provide moderate performance, while deep learning-based approaches achieve higher classification capability. Our proposed Gaussian-CNN achieves the best performance with an accuracy of 0.92, outperforming CNN (0.90), SVM (0.87), and LR (0.84). We observe that Gaussian filtering improves input feature quality by reducing noise and enhancing representation stability before convolution. This study concludes that integrating Gaussian preprocessing with CNN architecture significantly improves classification performance in psychological behavior analysis. We demonstrate that our proposed method enhances feature robustness and predictive accuracy, making it suitable for mental health monitoring and early detection in university environments.