Explainable machine learning framework with Granger-based temporal predictive validation for monitoring fiscal risk in national health insurance expenditure
- 1School of Computer Software, Daegu Catholic University, Gyeongsangbuk-do, South Korea.
Abstract
Rising health insurance expenditure (HIE) poses fiscal risks and challenges to the sustainability of public insurance systems. Despite growing interest in data-driven analysis, empirical evidence on the associations between macroeconomic, social, and financial conditions and national HIE remains limited, particularly across provider types. This study introduces an explainable machine-learning framework with temporal predictive validation to analyze and forecast HIE using 60 macro-financial and socioeconomic indicators. The framework is applied to three monthly expenditure series from Korea’s National Health Insurance system (2000–2023). Predictive performance and interpretability are assessed using two univariate autoregressive baselines, three regularized linear indicator-based benchmarks, and six nonlinear indicator-based ensemble models. LightGBM serves as the primary explanatory model, and SHapley Additive exPlanations (SHAP) quantifies indicator contributions and captures autoregressive persistence. Beyond associational insights, temporal predictive validation using lagged regression and Granger-type temporal predictive tests examines, within a prespecified exploratory lag window, whether SHAP-identified indicators show statistically supported exploratory lagged monitoring relevance. These results are interpreted as hypothesis-generating temporal monitoring evidence rather than as evidence of structural causal effects. Although persistence-based baselines achieved the strongest short-term forecasting accuracy, indicator-based models provided model-conditional explanatory information beyond autoregressive controls. The results indicate that shared macroeconomic pressures and provider-specific factors are associated with HIE predictions. The cost-of-living index is consistently associated with expenditure patterns across categories, whereas domestic demand and external conditions indicate provider-specific associations for medical institutions and pharmacies. Only a subset of predictors retained exploratory lagged monitoring support, underscoring the distinction between model-based predictive importance and hypothesis-generating monitoring signals. This integrated framework may inform transparent fiscal risk monitoring hypotheses and contextual review of public HIE.
Related
- A Multi-Criteria Geostatistical Framework Integrating Centrality Measures and Satellite Validation for PM2.5 Monitoring Network Expansion in Upper Northern Thailand
- Explainable spatiotemporal Graph Neural Networks: Multi-City PM2.5 forecasting in Bangladesh with geographic graph topology validation
- Multimodal computer vision and machine learning for predictive monitoring of planar-flow casting of amorphous steel foils
- A Shape-Adaptive Gamma-Mixture Power Shanker Distribution: Boundary-Aware Inference, Censored Regression, and Predictive Validation
- Model Predictive Control for Dynamic Hydraulic Balancing in Building Radiator Heating Systems: Modeling, Design, and Experimental Validation
BibTeX
@article{lee2026explainable,
title = {{Explainable machine learning framework with Granger-based temporal predictive validation for monitoring fiscal risk in national health insurance expenditure}},
author = {Gwangsu Lee},
journal = {PLOS One},
year = {2026},
month = oct,
doi = {10.1371/journal.pone.0361904},
url = {https://doi.org/10.1371/journal.pone.0361904},
note = {CC BY},
}
RIS
TY - JOUR
TI - Explainable machine learning framework with Granger-based temporal predictive validation for monitoring fiscal risk in national health insurance expenditure
AU - Gwangsu Lee
PY - 2026
DA - 2026/10/29
JO - PLOS One
DO - 10.1371/journal.pone.0361904
UR - https://doi.org/10.1371/journal.pone.0361904
AB - Rising health insurance expenditure (HIE) poses fiscal risks and challenges to the sustainability of public insurance systems. Despite growing interest in data-driven analysis, empirical evidence on the associations between macroeconomic, social, and financial conditions and national HIE remains limited, particularly across provider types. This study introduces an explainable machine-learning framework with temporal predictive validation to analyze and forecast HIE using 60 macro-financial and socioeconomic indicators. The framework is applied to three monthly expenditure series from Korea’s National Health Insurance system (2000–2023). Predictive performance and interpretability are assessed using two univariate autoregressive baselines, three regularized linear indicator-based benchmarks, and six nonlinear indicator-based ensemble models. LightGBM serves as the primary explanatory model, and SHapley Additive exPlanations (SHAP) quantifies indicator contributions and captures autoregressive persistence. Beyond associational insights, temporal predictive validation using lagged regression and Granger-type temporal predictive tests examines, within a prespecified exploratory lag window, whether SHAP-identified indicators show statistically supported exploratory lagged monitoring relevance. These results are interpreted as hypothesis-generating temporal monitoring evidence rather than as evidence of structural causal effects. Although persistence-based baselines achieved the strongest short-term forecasting accuracy, indicator-based models provided model-conditional explanatory information beyond autoregressive controls. The results indicate that shared macroeconomic pressures and provider-specific factors are associated with HIE predictions. The cost-of-living index is consistently associated with expenditure patterns across categories, whereas domestic demand and external conditions indicate provider-specific associations for medical institutions and pharmacies. Only a subset of predictors retained exploratory lagged monitoring support, underscoring the distinction between model-based predictive importance and hypothesis-generating monitoring signals. This integrated framework may inform transparent fiscal risk monitoring hypotheses and contextual review of public HIE.
SN - 1932-6203
ER -