Explainable Machine Learning in the Prediction of Depression.

Area of Science:

• Public Health
• Computational Psychiatry
• Machine Learning in Healthcare

Background:

• Depression is a significant global public health concern, with both genetic and environmental risk factors.
• Identifying modifiable environmental factors is crucial for reducing depression prevalence.
• This study focuses on environmental and sociodemographic predictors of depression.

Purpose of the Study:

• To investigate the association between depression and various sociodemographic, lifestyle, and health factors in Thrace, Greece.
• To compare the performance of four machine learning (ML) models in predicting depression.
• To identify key predictors of depression using feature selection and interpretation techniques.

Main Methods:

• A cross-sectional, questionnaire-based study was conducted on a sample from Thrace, Greece.
• Four ML models were employed: logistic regression (LR), support vector machine (SVM), XGBoost, and neural networks (NNs).
• A genetic algorithm (GA) was used for feature selection, and SHAP (SHapley Additive exPlanations) for model interpretation.

Main Results:

• XGBoost achieved the highest prediction accuracy (97.83%) for depression, followed closely by NNs (97.02%).
• The GA identified 15 significant risk factors utilized by the XGBoost model.
• SHAP analysis highlighted anxiety, education level, alcohol consumption, and body mass index as primary predictors.

Conclusions:

• Machine learning models, particularly XGBoost, show high efficacy in predicting depression based on identified risk factors.
• Findings support the development of personalized public health interventions and clinical strategies for mental well-being.
• Future research should focus on larger datasets to enhance model accuracy for early detection and personalized mental healthcare.