This research evaluates how structured reasoning methods can enhance large language models' abilities to predict mental health conditions from online text data.

• Chain-of-Thought, Self-Consistency, and Tree-of-Thought techniques significantly improved classification accuracy
• Tested across multiple Reddit-sourced mental health datasets
• Enhanced interpretability and robustness compared to traditional classification approaches
• Demonstrates potential clinical applications for early mental health screening

Why it matters: These approaches could enable more reliable, transparent AI-assisted mental health monitoring systems while providing clinicians with meaningful reasoning paths behind predictions, potentially transforming early intervention practices.

Cognitive-Mental-LLM: Evaluating Reasoning in Large Language Models for Mental Health Prediction via Online Text