Speaker
Description
The growing threat of environmental degradation has prompted renewed interest in the relationship between pollution and public health outcomes. Among environmental indicators, carbon dioxide (CO₂) emissions play a central role, both as a major contributor to climate change and as a proxy for the intensity of economic activities associated with health risks. While numerous studies have examined the impact of aggregate CO₂ emissions on health expenditure using traditional econometric models, relatively few have addressed the sectoral origin of emissions or adopted non-linear modeling frameworks capable of capturing complex interactions among environmental, economic, and health-related variables.
In this context, the present study aims to investigate the relationship between CO₂ emissions, broken down by economic sector (industry, agriculture, transportation, and electricity production), and per capita health expenditure in Morocco, using an innovative approach based on machine learning. Unlike traditional econometric methods, which rely on assumptions of linearity and strict model specification, machine learning algorithms can capture complex, nonlinear relationships between explanatory variables and target variables. More specifically, this work uses the Extreme Gradient Boosting (XGBoost) model, recognized for its performance in processing medium-sized datasets and its ability to model interactions and nonlinear effects.
The dataset consists of macroeconomic and environmental variables drawn from international databases such as the World Bank (WDI). The dependent variable is per capita health expenditure, while the main explanatory variables are CO₂ emissions from four sectors (industry, agriculture, transport, and electricity generation), along with GDP per capita as a control for income-related effects. The analysis covers the period from 2000 to 2024.
By leveraging the XGBoost model, this study not only trains a predictive framework but also evaluates the relative importance of each variable in explaining health expenditures. This approach is particularly robust in the presence of multicollinearity, missing data, and complex interactions that are typically difficult to model with traditional linear frameworks. Moreover, it enables the detection of threshold effects and sector-specific impacts, offering a richer understanding of the environmental-health nexus.
Empirical results demonstrate that CO₂ emissions from the transport sector are the most significant driver of health expenditure, followed by emissions from electricity generation and industry. In contrast, agricultural emissions and GDP per capita show weaker and more dispersed impacts. The model performs exceptionally well, achieving an R² of 0.998, with a Root Mean Squared Error (RMSE) of 4.85 and a Mean Absolute Error (MAE) of 3.71. These results confirm the strong explanatory power of XGBoost and highlight that pollution from sectors such as transport and power generation, which are closely linked to urbanization, plays a major role in driving public health expenditure.
This project is part of a multidisciplinary perspective at the intersection of health economics, environmental economics, and data science. It aims to contribute to the literature by proposing an empirical framework adapted to the analysis of the environmental and social sustainability of development trajectories. The challenge is also methodological: by introducing machine learning into economic analysis applied to health, this work highlights new possibilities for the diagnosis, forecasting, and evaluation of public policies.
In terms of policy implications, this study supports the case for sector-specific environmental regulation and health financing strategies that account for the externalities generated by pollution. For countries in transition or experiencing rapid urbanization, such as Morocco, this research offers a timely contribution to designing integrated policies that support both environmental sustainability and the long-term resilience of health systems.
