Predicting ecological footprint based on global macro indicators in G-20 countries using machine learning approaches.

Area of Science:

• Environmental economics
• Ecological footprint analysis
• Machine learning for environmental science

Background:

• Ecological impact (EF) is a critical global economic issue, influenced by human activities like land use, infrastructure, and resource consumption.
• Accurate prediction of EF is essential for sustainable development and policy-making.

Purpose of the Study:

• To compare the predictive capabilities of penalized regression (PR) methods (Ridge, Lasso, Elastic Net) and artificial neural networks (ANN) for ecological footprint (EF) indices.
• To evaluate these models using global database data for G-20 countries from 1999-2018.

Main Methods:

• Utilized penalized regression (Ridge, Lasso, Elastic Net) and artificial neural networks (ANN) for EF index prediction.
• Employed 10-fold cross-validation to assess predictive performance and tune model parameters.
• Compared model accuracy against Ordinary Least Squares (OLS) regression.

Main Results:

• Penalized regression methods showed moderate improvement over linear regression, with Elastic Net selecting more variables than Lasso.
• Lasso regression offered better predictive performance among PR models despite selecting fewer indicators.
• Artificial neural networks (ANN) significantly outperformed both PR and OLS, exhibiting higher R-squared values and lower RMSE, indicating greater predictive accuracy.

Conclusions:

• While PR methods offer valuable variable selection and model interpretability, ANN models provide superior predictive accuracy for ecological footprint indicators.
• ANN models are recommended for accurate EF predictions in G-20 countries, aiding environmental and economic policy.