박민영, 강정은, 이송주, 황원실 (2025). Explaining Nonlinear Relationships Between Community Factors and Heat-Related Illnesses Using Machine Learning. 20th APRU MULTI-HAZARDS SYMPOSIUM AND CONFERENCE 2025. (2025. 11. 25. - 2025. 11. 29.)  

[Abstract] 

  The aim of this study is to analyze the nonlinear relationships between regional characteristics and the occurrence of heat-related illnesses, providing valuable insights for the development of heatwave-related policies. Machine learning algorithms and SHAP values, known for their effectiveness in analyzing nonlinear relationships, were utilized. Regional characteristic variables were constructed based on the three components of risk assessment―hazard, and vulnerability―and the model was trained using a 7-year dataset from 2013 to 2019. SHAP values were calculated for the algorithm that demonstrated the best performance. The analysis revealed that apparent temperature, the proportion of elderly residents, and water accessibility had a positive impact on the number of heat-related illness cases per administrative district population as their values increased. Conversely, the proportion of urbanized areas and the proportion of people with disabilities were found to have a negative impact. Notably, the apparent temperature showed a sharp positive effect at 36°C, while the number of outdoor workers exhibited a significant positive impact within specific high-value ranges. Additionally, the accessibility of heatwave shelters positively influenced the occurrence of heat-related illnesses in administrative districts where a higher concentration of shelters was located around 250 meters. These findings support more effective, locally tailored policies. 

사사 : 기후탄력성, 환경보건센터과제