The construction sector is one of the top contributors to the world's carbon footprint, owing to its energy requirements and resource consumption. This study aims to understand the impact of Machine Learning (ML) on estimating and assessing the carbon footprint of construction activities using multi-sector carbon datasets. Key components such as energy usage, emissions at the sector level, GDP, population, and the proportion of renewable energy were used to fine-tune and assess multiple regression-based ML algorithms. Six baseline models were created: Linear Regression, Ridge, Lasso, Support Vector Regression (SVR), Decision Tree, and Random Forest, as well as advanced ensemble methods XGBoost and LightGBM. Additional feature engineering was utilized to develop normalized emission ratios, such as per capita and per GDP. A broad range of evaluation indicators was used, including R² score, RMSE, MAE, MAPE, MSLE, median absolute error, and explained variance. The outcome indicated that traditional linear models were more predictable (R² ≈ -0.007, RMSE ≈ 85.97) while tree-based Random Forest models struggled (R² ≈ -0.034, RMSE ≈ 87.12), which means none of the parallel models outperformed the emission variance. XGBoost and LightGBM achieved similar yields; xGBoost earned R² = -0.235, RMSE = 95.20, illustrating that a model based on complex, high-dimensional environmental data is difficult to construct. In a hypothetical situation where the use of renewable energy sources was increased by 20%, most models still forecasted only slight emission reductions (for example, Random Forest: Change from 152.92 to 152.87 metric tons). SHapley Additive exPlanations (SHAP) explainability pointed to energy demand, industrial CO₂, and proportion of renewables as the main contributors to emissions. Further cluster analysis revealed distinct emission profiles by region, which can inform focused environmental policy. This research analyzes the possibility of applying machine learning to discover structural features in carbon emissions and assesses the impact of renewable energy policies in the construction industry. The study also stresses the role of explainability and feature engineering on environmental simulations.
A Sun, study studied this question.