An Ensemble Classification Model for Depression Based on Wearable Device Sleep Data.

Area of Science:

• Digital Health
• Computational Psychiatry
• Machine Learning

Background:

• Depression is a common mental disorder often accompanied by sleep disturbances.
• Wearable devices are increasingly used for sleep tracking, presenting opportunities for health monitoring.
• Existing depression detection methods often struggle with missing data and complex data utilization.

Purpose of the Study:

• To develop an intelligent and economical depression detection system using wearable sleep data.
• To address the challenge of missing data in wearable device datasets.
• To improve the robustness and generalizability of depression identification models.

Main Methods:

• An improved Generative Adversarial Imputation Network (GAIN) was developed to handle missing sleep data, showing a 28.56% improvement in Mean Absolute Error (MAE).
• An ensemble classification model for depression (ECD) was constructed by combining Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Logistic Regression (LR), Case-Based Reasoning (CBR), and Decision Tree (DT) models.
• A voting mechanism was integrated into the ensemble model to enhance noise immunity.

Main Results:

• The improved GAIN method demonstrated superior performance in interpolating missing values compared to the original GAIN approach.
• The ensemble classification model achieved high performance in depression detection, with a precision of 92.55% and a recall of 91.89%.
• The proposed method shows significant efficiency in automatic depression detection using sleep data.

Conclusions:

• The developed ensemble classification model (ECD) offers an effective and robust solution for depression detection using wearable sleep data.
• The improved data imputation technique addresses a critical limitation of wearable devices, enhancing data usability.
• This approach provides a promising, intelligent, and economical avenue for widespread mental health screening and monitoring.