AI-Driven Multiclass Mental Health Prediction Using NLP and Deep Learning Models

Abstract

The rapid development of social media sites has resulted in massive volumes of textual data that has the potential to expose the psychological condition of users. This paper presents a framework of AI-based mental health study on social media posts and is aimed at identifying mental conditions at an initial stage, specifically depression. The system combines machine learning models and deep learning with natural language processing techniques to extract meaning from linguistic and contextual features. Tokenization, normalization, and feature extraction by use of TF-IDF and word embeddings are used to preprocess texts. Various models (such as Support Vector Machine, Artificial Neural Network, Long Short-Term Memory, and transformer-based BERT) are used to perform classification. The experiments show that the deep learning methods outperform the traditional models with up to 92-99% accuracy with better precision and F1-score. Specifically, transformer-based models are better because they can learn contextual dependencies and semantic relationships in text. The results affirm that social media data can be efficiently used to do early mental health screening. The suggested framework allows scalable and real-time monitoring frameworks and allows the provision of timely intervention. Nevertheless, issues like data privacy, ethical considerations as well as imbalance of data need to be mitigated in a practical implementation.