Explainable Machine Learning for Heat-Related Illness Prediction: An XGBoost-SHAP Approach Using Korean...

Area of Science:

• Environmental health
• Climate change adaptation
• Public health informatics

Background:

• Climate change is increasing the frequency of heat-related illnesses (HRIs), posing significant public health challenges, especially in urban areas.
• Accurate prediction of HRI risk is crucial for timely public health interventions and planning.

Purpose of the Study:

• To develop and validate an explainable machine learning (ML) model for predicting daily HRI risk.
• To identify key meteorological factors contributing to HRI risk using explainable AI (XAI).
• To support the development of localized early warning systems for climate-sensitive diseases.

Main Methods:

• Applied eXtreme Gradient Boosting (XGBoost) to model HRI occurrence based on daily meteorological data (temperature, humidity, solar radiation, wind speed, precipitation).
• Utilized Shapley Additive exPlanations (SHAP) for model interpretability, identifying key risk drivers.
• Validated model performance using historical data and time-series comparisons.

Main Results:

• The ML model demonstrated strong predictive accuracy for HRI risk (AUC = 0.895).
• Mean daily temperature, solar radiation, and minimum temperature were identified as the most significant contributors to HRI risk.
• The model effectively predicted HRI occurrences in real-world settings.

Conclusions:

• Explainable AI (XAI) offers a powerful approach for localized health-risk forecasting.
• The developed model provides a data-driven foundation for proactive public health planning against escalating urban heat risks.
• Findings support the implementation of targeted early warning systems to mitigate climate-sensitive disease burdens.