Federated AI for Mental Health: A Privacy-Preserving GAN-LLM Framework for Risk Prediction and Early Detection

DOI

10.2991/978-94-6239-705-7_10How to use a DOI?

Keywords

Federated Learning (FL); Generative Adversarial Networks (GANs); Large Language Models (LLMs); Mental Health Risk Prediction; Privacy-Preserving AI

Abstract

Mental health disorders affect nearly 30% of South Africa’s population, yet access to adequate care remains significantly limited due to socioeconomic barriers, stigma, and resource shortages. This study presents Federated GANS-LLM, a novel hybrid framework that combines federated learning (FL), generative adversarial networks (GANs), and large language models (LLMs) to improve mental health risk prediction, facilitate early detection, and enhance patient-centered care while maintaining strict data privacy compliance. The proposed framework is built on a multi-layered architecture comprising three key components. The Client-Side Federated Learning layer enables healthcare institutions to process patient data locally, transmitting only encrypted model updates to ensure privacy protection. The GAN-Based Data Augmentation layer includes a centralized GAN module enhanced with variational autoencoders (VAE) to generate high-quality synthetic data, addressing data sparsity and heterogeneity to improve model robustness. The ClinicalBERT-Driven Risk Assessment layer employs an advanced LLM that analyzes contextual indicators, sentiment, and linguistic markers to assess mental health risks with high interpretability. By assessing the technical feasibility, ethical considerations, and real-world impact of this approach, the research contributes to the advancement of scalable, privacy-preserving AI solutions for addressing global mental health challenges. This work also demonstrates the transformative potential of AI-driven, privacy-focused innovations in healthcare, positioning federated learning as a viable approach to bridging the mental health treatment gap worldwide.

Copyright

© 2026 The Author(s)

Open Access

Open Access This chapter is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International License (http://creativecommons.org/licenses/by-nc/4.0/), which permits any noncommercial use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.

Download article (PDF)

TY  - CONF
AU  - N. P. Thothela
AU  - A. Bagula
PY  - 2026
DA  - 2026/06/13
TI  - Federated AI for Mental Health: A Privacy-Preserving GAN-LLM Framework for Risk Prediction and Early Detection
BT  - Proceedings of the FIREtalk Conference - Research on FIRE! (research-on-fire 2025)
PB  - Atlantis Press
SP  - 136
EP  - 155
SN  - 2352-5398
UR  - https://doi.org/10.2991/978-94-6239-705-7_10
DO  - 10.2991/978-94-6239-705-7_10
ID  - Thothela2026
ER  -