AI-based analysis of key predictors of CO2 emissions across income groups

Abstract

This study examines how macroeconomic, energy, demographic, and trade factors are associated with CO 2 emissions across country income groups using World Bank World development indicators (WDI) data (1990–2023). A set of deep learning models (multilayer perceptron (MLP), recurrent neural network (RNN), long short-term memory (LSTM)) is evaluated to capture non-linear patterns, and the MLP yields the best predictive accuracy. To interpret the main predictors underlying model performance, random forest (RF) feature-importance and correlation analyses are conducted, revealing systematic heterogeneity across development stages: In lower-income countries, emission patterns are more strongly linked to the energy mix and natural-resource endowments, whereas in middle- and high-income economies they are increasingly associated with trade exposure, fuel dependence, industrial activity, and demographic pressures. These findings should be interpreted as predictive associations rather than causal effects and support the design of differentiated mitigation policies aligned with countries' structural characteristics, rather than uniform global targets, to improve both effectiveness and equity in emission reduction strategies.