Explainable AI for Depression Detection and Severity Classification From Activity Data: Development and Evaluation...

Area of Science:

• Wearable technology and digital health
• Machine learning in mental health
• Explainable Artificial Intelligence (XAI)

Background:

• Depression affects 280 million globally, often undiagnosed, highlighting the need for objective detection methods.
• Wearable devices offer continuous activity monitoring for data-driven depression assessment.
• Existing machine learning models struggle with depression subtype classification and lack clinical transparency.

Purpose of the Study:

• Develop and evaluate an interpretable machine learning framework for depression detection and severity classification.
• Utilize wearable actigraphy data to overcome challenges of imbalanced datasets and model transparency.
• Enhance clinical acceptance through explainable AI in mental health diagnostics.

Main Methods:

• Applied Adaptive Synthetic Sampling (ADASYN) to address class imbalance in the Depresjon dataset.
• Extracted statistical features (e.g., power spectral density mean, autocorrelation) and demographic data from activity logs.
• Evaluated five machine learning algorithms (logistic regression, SVM, random forest, XGBoost, neural networks) using standard performance metrics and employed SHAP and LIME for interpretability.

Main Results:

• XGBoost demonstrated superior performance with 84.94% accuracy for binary classification and 85.91% for multiclass severity.
• Shapley Additive Explanations (SHAP) and Local Interpretable Model-agnostic Explanations (LIME) identified power spectral density mean, age, and autocorrelation as significant predictors.
• Findings underscore the role of circadian rhythm disruptions in depression, as indicated by activity pattern analysis.

Conclusions:

• The interpretable framework effectively distinguishes between depressed and non-depressed individuals and classifies depression severity (mild vs. moderate).
• Integration of SHAP and LIME provides transparent, clinically relevant insights into prediction drivers.
• Explainable AI holds significant potential for improving early depression detection and intervention strategies in mental healthcare.