Presentation Information
[POS-26]Development of a Region-Specific Warning System Based on Machine Learning Methods of Heat-Related Illnesses
*Jeonghwa Seo1, Geunsoo Jang2, Giphil Cho3, Hiroshi Nishiura4, Hyojung Lee1 (1. Department of Statistics, Kyungpook National University, Daegu, 41566 (Korea), 2. Nonlinear Dynamics & Mathematical Application Center, Kyungpook National University, Daegu, 41566 (Korea), 3. Department of Electronic and AI System Engineering, Kangwon National University, Samcheok-si, Gangwon-do, 25913 (Korea), 4. Kyoto University School of Public Health, Yoshida-Konoe, Sakyo, Kyoto, 606-8601 (Japan))
Keywords:
Heat-related illness,Machine learning method,Region-specific prediction model,Warning system,Heatwave
The increasing frequency of heatwaves driven by climate change has significantly heightened the risk of heat-related illnesses. However, there remains no universally accepted definition of a heatwave. In Korea, the Korea Meteorological Administration (KMA) defines a heatwave as two consecutive days when the apparent temperature exceeds 33°C. However, this general criterion does not adequately reflect the diverse climatic and socioeconomic characteristics across regions.
This study aimed to develop region-specific prediction models to establish an effective early warning system for heat-related illnesses. We utilized data collected between 2016 and 2023, including heat-related illness cases, meteorological factors, and regional characteristics. The prediction accuracy of Linear Regression (LR), Random Forest (RF), eXtreme Gradient Boosting (XGB), and Long Short-Term Memory (LSTM) models were compared to select the optimal predictive model.
In addition to the Wet Bulb Globe Temperature (WBGT), which is widely used for assessing heat-related illness risk, we incorporated region-specific variables and indicators reflecting local heat acclimatization into the models. After identifying the best performing model, we predicted heat-related illness cases for each region and established warning thresholds, based on which we proposed a new heat warning system.
Unlike the current warning system that relies solely on apparent temperature, our proposed method provides a customized approach based on regional characteristics. This enhanced system is expected to improve public health outcomes by effectively managing the risks associated with heat-related illnesses.
This study aimed to develop region-specific prediction models to establish an effective early warning system for heat-related illnesses. We utilized data collected between 2016 and 2023, including heat-related illness cases, meteorological factors, and regional characteristics. The prediction accuracy of Linear Regression (LR), Random Forest (RF), eXtreme Gradient Boosting (XGB), and Long Short-Term Memory (LSTM) models were compared to select the optimal predictive model.
In addition to the Wet Bulb Globe Temperature (WBGT), which is widely used for assessing heat-related illness risk, we incorporated region-specific variables and indicators reflecting local heat acclimatization into the models. After identifying the best performing model, we predicted heat-related illness cases for each region and established warning thresholds, based on which we proposed a new heat warning system.
Unlike the current warning system that relies solely on apparent temperature, our proposed method provides a customized approach based on regional characteristics. This enhanced system is expected to improve public health outcomes by effectively managing the risks associated with heat-related illnesses.