Factors influencing global life expectancy across different income levels using Shapley value decomposition

The convergence of life expectancy across nations is a pivotal subject of inquiry in global health economics. Traditional econometric approaches have difficulty precisely allocating the relative contributions of multiple interacting variables, especially in nonlinear or complex models, which the authors identify as the primary limitation in existing literature. This paper addresses this limitation by using the Shapley value decomposition (SHAP) alongside a Light Gradient Boosting Machine (LGBM) to examine the key factors influencing life expectancy at birth across countries with different income levels. The LGBM model significantly outperformed a traditional linear regression model (R 2 of 0.901 vs. 0.716; MAPE of 2.08% vs. 3.38%). The SHAP analysis revealed that importance of predictive factor varies considerably by income level. For instance, the fertility rate is the most dominant factor in low-income countries, while health expenditure per capita is the most influential variable in high-income regions. The study concludes that model-informed policy considerations must be tailored to the specific economic context of a country, prioritizing basic infrastructure improvements in lower-income nations while focusing on advanced healthcare investments in wealthier ones.