Assessing Future Heatwave-Related Mortality in Greece Using Advanced Machine Learning and Climate Projections

Ilias Petrou, Pavlos Kassomenos, Nikolaos Kyriazis

Climate change has intensified the frequency and severity of heatwaves globally, posing significant public health risks, particularly in Mediterranean countries such as Greece, where rising temperatures coincide with vulnerable aging populations. This study develops a machine learning framework employing XGBoost models to predict monthly heatwave-attributable mortality from cardiovascular and respiratory diseases across Greek regions, stratified by age groups. Using high-resolution climate projections under RCP4.5 and RCP8.5 scenarios, the models integrate meteorological and demographic data to capture complex nonlinear relationships and regional heterogeneity. Model performance was rigorously validated with a temporally held-out dataset, demonstrating high predictive accuracy (R2 > 0.96). Projections indicate a sharp increase in elderly mortality due to heat exposure by mid-century, with marked geographic disparities emphasizing urban centers like Attica. This work advances prior studies by incorporating detailed spatial and demographic stratification and applying robust machine learning techniques beyond traditional statistical approaches. The model offers a valuable tool for public health planning and climate adaptation in Greece and similar Mediterranean contexts. Our findings highlight the urgent need for targeted mitigation strategies to address the growing burden of heatwave-related mortality under changing climate conditions.