Development of a Machine Learning-Based Model for Classifying Depression Using Physiological and Psychological...

Area of Science:

• Biomedical Informatics
• Psychiatry
• Machine Learning

Background:

• Depression classification relies on psychological indicators, but physiological markers are underexplored.
• Salivary cortisol and dehydroepiandrosterone (DHEA) offer potential biomarkers for depression.

Purpose of the Study:

• Develop and evaluate machine learning models for depression symptom classification.
• Assess the contribution of salivary cortisol and DHEA to classification accuracy.
• Compare multi-class (four degrees) and binary (normal vs. depression) classification models.

Main Methods:

• Utilized psychological indicators and salivary cortisol/DHEA levels from 368 training and 92 testing individuals.
• Developed multi-class and binary classification models, evaluating performance with and without physiological data.
• Employed k-fold cross-validation, class weight balancing, and grid search for model robustness.

Main Results:

• The multi-class model achieved 85.9% accuracy with physiological indicators (vs. 76.1% without).
• The binary classification model reached 97.8% accuracy, irrespective of physiological indicator inclusion.
• Selected 24 non-demographic variables for model development.

Conclusions:

• Machine learning models demonstrate high accuracy for depression classification.
• Integrating psychological and physiological markers, including salivary hormones, enhances classification performance.
• Further clinical trials are needed to validate model reliability and clinical utility.