Multiclass Mental Illness Prediction Using Lstm And Natural Language Processing

Abstract

 Mental health disorders have become a significant global concern, affecting millions of individuals worldwide. With the increasing usage of social media platforms, users often express their emotions, thoughts, and psychological conditions through textual content. Detecting mental illness from such data is a complex task due to informal language, emotional depth, and metaphorical expressions.

This research presents a hybrid deep learning framework for multiclass mental illness prediction using Natural Language Processing (NLP). The system integrates domain-specific transformer models such as MentalBERT and MelBERT with Convolutional Neural Networks (CNN) and Bidirectional Long Short-Term Memory (BiLSTM) networks. MentalBERT captures contextual mental health-related features, while MelBERT identifies metaphorical language patterns.

The model processes social media text data and classifies it into multiple mental health conditions such as depression, anxiety, PTSD, and bipolar disorder. Experimental results show improved accuracy and performance compared to traditional machine learning methods.

This work contributes to early detection and mental health awareness through intelligent text analysis systems.