This study evaluates the predictive accuracy of Regression Trees (RTrees) and Adaptive Neuro-Fuzzy Inference Systems (ANFIS) for estimating the carbon footprint in residential construction projects. The results indicate that the ANFIS significantly outperforms the RTrees in predictive accuracy, achieving a reduction in Root Mean Square Error (RMSE) by 84.3% in the production stage (from 0.53174 to 0.08346) and by 40.4% in the operational stage (from 0.13865 to 0.08265). These improvements underscore the effectiveness of the ANFIS in capturing complex nonlinear relationships in carbon footprint data. Despite its superior accuracy, the ANFIS exhibits higher computational costs, requiring an average training time of 76.2 s, compared to 12.4 s for the RTrees. These findings highlight the trade-offs between accuracy and computational efficiency, providing valuable insights for selecting machine learning models in sustainable construction. The study concludes that integrating hybrid approaches or ensemble learning could further enhance predictive performance while maintaining efficiency.