DeepSentiMind: An Ensemble Intelligence Framework for Social Media-Based Mental Health Prediction

Abstract

Sentiment analysis plays a vital role in identifying emotional patterns expressed through social media content, enabling early recognition of mental health concerns. DeepSentiMind presents an intelligent ensemble framework for social media-based mental health prediction using natural language processing and machine learning. The framework preprocesses textual data, extracts meaningful linguistic features, and combines Support Vector Machine, Convolutional Neural Network, and Logistic Regression through soft voting to improve classification reliability. Publicly available Reddit datasets support experimental evaluation, while TF-IDF representations and pretrained GloVe embeddings strengthen textual understanding. Performance comparison demonstrates superior predictive capability, achieving accuracy exceeding ninety-five percent compared with individual classifiers. A Django-based web application provides real-time prediction, user interaction, dataset management, model retraining, and visualization dashboards for performance monitoring. The proposed approach supports timely identification of depression indicators from user-generated posts while maintaining a scalable architecture suitable for research and practical screening environments. Experimental findings indicate that ensemble learning reduces individual model limitations and improves prediction consistency. The system offers separate interfaces for users and administrators, enabling secure data handling, historical result tracking, and accuracy visualization. Future enhancements include multilingual support, transformer-based models, multimodal analysis, continuous monitoring, and clinical validation to improve robustness, generalization, usability, reliability, deployment readiness, future research.