Comparative Analysis of Machine Learning Algorithms for Mental Health Prediction Using Social Media Data
Year: 2026
Type: preprint
Abstract: This paper presents a comparative analysis of five machine learning algorithms — Logistic Regression, Naive Bayes, Support Vector Machine (SVM), Random Forest, and XGBoost — for automated mental health prediction using social media text data. Using a combined dataset of 26,945 posts from Reddit and Twitter, a complete NLP pipeline was developed including tokenization, lemmatization, stop-word removal, and TF-IDF feature vectorization. XGBoost achieved the highest accuracy of 93.2% with an AUC-ROC of 0.967. The study includes confusion matrix analysis, ROC curves, cross-validation stability testing, SHAP-based explainability, and a discussion of ethical considerations including data privacy, algorithmic bias, and responsible AI. Limitations and future directions including BERT, Federated Learning, and multimodal analysis are outlined
Author: Diya Prithiani
Language: English
Topic: Mental Health via Writing
Subfield: Social Psychology
Field: Psychology
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W7164143061
Cites:
Cited by:
Related to: