A Multi-Method Machine Learning Analysis of Sleep Disturbances' Determinants During COVID-19.

Area of Science:

• Computational neuroscience
• Health informatics
• Machine learning applications in healthcare

Background:

• Sleep disturbances represent a significant public health concern.
• The scientific community is actively exploring machine learning (ML) for predicting sleep determinants.
• Understanding the factors influencing sleep is crucial for developing effective interventions.

Purpose of the Study:

• To evaluate the predictive accuracy of various machine learning algorithms for sleep problems.
• To identify key predictors contributing to sleep disturbances using feature selection techniques.
• To leverage Explainable AI (XAI) for interpreting ML model predictions.

Main Methods:

• Utilized a publicly available dataset for sleep disturbance analysis.
• Applied multiple feature selection techniques to pinpoint influential predictors.
• Employed Explainable AI (XAI) methods, specifically SHAP values, to interpret predictor impact.

Main Results:

• The AdaBoost algorithm demonstrated superior performance with 71.27% accuracy.
• Sleep quality was identified as the most dominant predictor of sleep disturbances (SHAP value: 0.01586).
• Feature importance analysis revealed key factors influencing sleep health predictions.

Conclusions:

• Explainable AI (XAI) methods, such as SHAP, significantly enhance the clinical utility of ML models for sleep health.
• These insights enable healthcare providers to design targeted interventions for improving patient sleep outcomes.
• The study underscores the potential of ML and XAI in advancing sleep medicine and public health strategies.