XGBoost PSO for Mental Health Prediction Using Social Media Analytics

View at publisher Summarise with VIRAT
Authors: Akshaya K, Arulanantham G, Suryarajan S
Cited by: 0
References: 14

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 disorders represent a critical global health challenge aligned with Sustainable Development Goal 3 (Good Health and Well-Being), requiring innovative detection and intervention strategies. This research work proposes a hybrid machine learning framework integrating XGBoost with Particle Swarm Optimization (PSO) for automated mental health disorder prediction from social media text data. The proposed XGBoost-PSO model achieves exceptional performance with accuracy of 99.56 %, precision of 99.57 %, recall of 99.56 %, and$\mathbf{F 1}$-score of 99.56 % on the test dataset. The PSO optimizer efficiently identifies optimal hyperparameters including learning rate of 0.0111, maximum depth of 4, minimum child weight of 8.20, subsample ratio of 0.7935, and column sample by tree of 0.8830, with 60 estimators. Feature importance analysis reveals depression-related keywords contribute 74.50 %, followed by anxiety terms at 5.92 %, demonstrating the model's ability to capture critical mental health indicators. ROC analysis yields area under curve of 0.998, while ECDF demonstrates 96.8 % of predictions within 0.05 absolute error margin. SHAP explainability analysis provides interpretable insights into prediction mechanisms, supporting clinical validation and trust in automated mental health screening systems contributing to SDG 3 targets.

Fields of Study & Subjects

Mental Health via Writing Artificial Intelligence in Healthcare Sentiment Analysis and Opinion Mining

Publication details

Journal / source
International Conference on Communication, Computing and Electronics Systems (ICCCES)
Pages
1484–1489
Publication date
21 January 2026
Language
EN