Continuous Monitoring of Mental Health through Streaming Machine Learning with Counterfactual Explanations

Area of Science:

• Artificial Intelligence in Healthcare
• Computational Psychiatry
• Machine Learning for Mental Health

Background:

• Mental health disorders like anxiety and depression affect many, yet a significant portion do not receive care.
• Existing automatic detection solutions and treatments have limitations in accessibility and real-time application.
• There is a need for interpretable and adaptable systems for timely mental health assessment.

Purpose of the Study:

• To develop an innovative framework for multi-label detection of anxiety and depression.
• To integrate counterfactual explanations for enhanced interpretability and user-centered results.
• To enable real-time processing of mental health data through a streaming approach.

Main Methods:

• Utilized large language models (LLMs) for feature extraction.
• Employed multi-label machine learning for accurate prediction of anxiety and depression.
• Implemented a streaming mode with sliding window techniques for temporal data management.
• Generated counterfactual explanations to ensure transparency and accountability.

Main Results:

• Achieved over 90% accuracy in detecting anxiety and depression labels.
• Demonstrated a low rate of misclassifications across all labels.
• Validated the framework's effectiveness in a real-time, streaming environment.

Conclusions:

• The proposed framework offers a timely and transparent decision-making tool for mental healthcare.
• Integration of counterfactual explanations enhances the interpretability and trustworthiness of AI in mental health.
• The system addresses critical gaps in real-time detection and care for mental health disorders.