Actigraphy-based step analysis for the detection of depressed mood: An explainable machine learning approach

Area of Science:

• Digital health
• Artificial intelligence in medicine
• Behavioral science

Background:

• Depression detection often relies on subjective assessments.
• Objective, continuous monitoring using wearable devices is needed.
• Actigraphy offers a promising tool for passive data collection.

Purpose of the Study:

• To develop an interpretable artificial intelligence (AI) model for detecting depressive symptoms using actigraphy data.
• To integrate statistically significant features into machine learning models for enhanced accuracy and explainability.
• To explore the utility of actigraphy-derived features in classifying depressive symptoms.

Main Methods:

• Actigraphy data from 3304 participants over one week were analyzed.
• Six machine learning models, including CatBoost and XGBoost, were trained on various activity and light exposure indicators.
• Shapley additive explanations (SHAP) were used for model interpretability, with analyses stratified by sex and gender.

Main Results:

• CatBoost and XGBoost showed the highest performance in predicting mild and moderate-to-severe depressive symptoms, respectively (AUROC 0.679-0.715).
• SHAP analysis identified low step counts and high activity during the least active 5-hour period (L5) as key predictors of depressive symptoms.
• Predictive accuracy was highest in older men, with AUROC values reaching 0.833.

Conclusions:

• Actigraphy-derived step count and temporal activity patterns are valuable for AI-driven depression classification.
• Explainable AI approaches are crucial for personalized mental health screening.
• Time-sensitive analysis of activity data holds significant potential for objective mental health assessment.