A Multimodal Deep Learning Framework for Early Detection of Mental Health Conditions from Social Media

View at publisher Summarise with VIRAT
Authors: Premkumar Duraisamy, Haneefdeen S, Lakshana SJ, Sadhana K
Cited by: 0
References: 21

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Mental health issues such as depression, anxiety, and stress are increasingly prevalent among individuals who actively engage on social media platforms. Posts and images shared online often contain subtle linguistic and visual cues that reflect users' psychological states. This research presents a multimodal deep learning framework that integrates advanced textual and visual analysis for mental health prediction. Textual data is processed through transformer-based models such as BERT and its optimized variant SqueezeBERT, while image-based emotional features are analyzed using computer vision techniques. Additionally, ensemble methods like Random Forest and XGBoost are employed to enhance classification performance and improve model interpretability, with SVM serving as a baseline comparator. Experimental results show that SqueezeBERT achieves the highest accuracy of$\mathbf{9 5. 4 8 \%}$and precision of$\mathbf{9 3. 1 \%}$, outperforming other algorithms in detecting early signs of mental distress. The study underscores that combining linguistic and visual modalities leads to more accurate and holistic detection of mental health risks. The proposed framework demonstrates strong potential for scalable, privacy-preserving, and ethical AI systems that support early mental health intervention and awareness in real-world social media environments.

Fields of Study & Subjects

Mental Health via Writing Emotion and Mood Recognition Sentiment Analysis and Opinion Mining

Publication details

Journal / source
International Conference on Communication, Computing and Electronics Systems (ICCCES)
Pages
1049–1054
Publication date
21 January 2026
Language
EN