Detection and Prediction of Future Mental Disorder from Social Media Data Using Machine Learning Ensemble Learning and Large Language Models

Abstract

With the exponential growth of user-generated content on social media, researchers are exploring new ways to extract meaningful patterns to understand public health trends—particularly mental health conditions. This paper investigates the detection and prediction of future mental disorders through social media analysis using a combination of machine learning, ensemble learning methods, and large language models (LLMs). The approach aims to identify behavioral and linguistic indicators of mental distress before clinical diagnosis or self-reporting. Ensemble methods such as Random Forests and Gradient Boosting are integrated with deep learning language models like BERT and RoBERTa to improve prediction accuracy. A diverse set of features including sentiment polarity, temporal posting patterns, and linguistic markers are extracted from social posts to train the models. The proposed system achieves high accuracy in predicting early warning signs of mental disorders, such as depression, anxiety, and PTSD, with explainability incorporated through SHAP values. This research offers a scalable, data-driven solution to assist clinicians and policymakers in early mental health intervention strategies.