A MACHINE LEARNING APPROACH BASED ON SYNTHETIC DATA FOR ESTIMATING CARBON FOOTPRINT IN LOGISTICS PROCESSES

Abstract

This study aims to estimate the carbon footprint arising from logistics processes using machine learning methods based on synthetic data and to identify the most suitable model for this purpose. Due to data access limitations, a multidimensional synthetic dataset was generated using variables like transport distance, cargo weight, transport mode, fuel type, traffic density, weather, and vehicle characteristics. Carbon emissions were calculated via an activity-based approach. The study comparatively evaluated six algorithms: Linear Regression, Random Forest, Extra Trees, Gradient Boosting, Support Vector Regression, and XGBoost. Findings indicate that tree-based ensemble learning models outperformed classical methods in predicting emissions. Among all models, XGBoost delivered the highest performance with an R² value of 0.9945 and minimal error rates. The results demonstrate that synthetic data provides an effective alternative for estimating carbon footprints in logistics process where access to real-world data is limited, and that it can contribute to sustainable logistics decision-making processes.