Multi-Class Mental Health Detection With LSTM and BilLSTM Models

DOI

https://www.doi.org/10.59256/ijire.20260702020

Published In

Volume 7, Issue 2 (2026)

Published Online

March-April 2026

Pages

152-164

Article Type

Original Article

Check for updates

Mental health disorders have become a major global concern, affecting millions of individuals worldwide. Traditional diagnostic methods are often time-consuming, reactive, and limited by accessibility and social stigma. With the rise of social media, users increasingly express their emotions through textual content, creating opportunities for early detection of mental health issues using computational techniques. This research focuses on developing a deep learning-based framework for multi-class mental health detection using Natural Language Processing (NLP). The study utilizes the Mental Distress (2026) dataset, which consists of social media text categorized into five classes: Depressed, Anxious, Frustrated, Suicidal, and Others. Unlike traditional binary classification approaches, this research addresses a more complex multi-class classification problem, where emotional states often overlap linguistically and semantically. Advanced preprocessing techniques, including text normalization, tokenization, stop word removal, and sequence padding, are applied. The text is converted into numerical form using embedding layers to capture semantic relationships. Two models, Long Short-Term Memory (LSTM) and Bidirectional LSTM (BiLSTM), are implemented and compared. The BiLSTM model, with its bidirectional learning capability, provides better contextual understanding. The models are evaluated using Stratified 5-Fold Cross-Validation to ensure reliability and robustness. Performance metrics such as accuracy, precision, recall, and F1-score are used for evaluation. Experimental results demonstrate that the BiLSTM model outperforms the LSTM model, achieving an accuracy of approximately 89%, compared to 86% for LSTM. The improvement is attributed to the model’s ability to capture bidirectional context and resolve lexical ambiguities in emotional expressions. This research demonstrates the effectiveness of deep learning techniques in accurately detecting mental health conditions from textual data and highlights their strong potential for real-world applications. These include early warning systems for identifying psychological distress, AI-driven mental health monitoring tools, and digital support platforms that can assist individuals and healthcare professionals in timely intervention and decision-making.