Predicting and Explaining Unmet Healthcare Needs
Geospatial Machine Learning for Predicting and Explaining Unmet Healthcare Needs Across Local Governments
공간정보 기반 머신러닝을 활용한 지방정부의 미충족의료율 예측과 설명
Authors: Dohyo Jeong
Abstract
Geospatial Machine Learning for Predicting and Explaining Unmet Healthcare Needs Across Local GovernmentsRegional disparities in healthcare access exacerbate health inequalities and pose a major challenge for local governments. This study predicted unmet healthcare needs using machine learning models incorporating characteristics of neighboring regions among local governments in South Korea. Explainable artificial intelligence was used to identify global and local predictors. Prediction models were developed using Random Forest and Extreme Gradient Boosting (XGBoost) and evaluated through year-by-year validation. The model incorporating neighboring regional characteristics achieved the best predictive performance. The number of physicians and the proportion of single-person households in neighboring regions were important predictors of unmet healthcare needs. Moreover, even among regions with relatively high predicted unmet healthcare needs, the major predictors and their contributions differed. These findings suggest that predicting unmet healthcare needs requires consideration of both local and neighboring regional characteristics. By identifying regions with high predicted unmet healthcare needs and explaining region-specific prediction factors, this study provides an analytical framework for local healthcare planning and evidence-based policymaking.
Key Words: unmet healthcare needs, healthcare vulnerability, machine learning, spatial analysis, local government