Development of an interpretable machine learning model for depression detection in individuals, suggesting effective classification strategies.
Key Points
XGBoost achieved an accuracy of 84.94% for detecting depression, which shows promising potential in clinical applications.
The study utilized the Depresjon dataset and applied Adaptive Synthetic Sampling to effectively address class imbalance challenges.
Features like power spectral density mean and age were identified as key predictors for depression severity, emphasizing their importance in detection.
Incorporating SHAP and LIME enhances model interpretability, indicating which features influence predictions most for better clinical transparency.