Prediction of Fuel Burn for Runway-to-Runway Commercial Flight Operations with Machine Learning

The mitigation of aviation's environmental impacts is crucial in enabling robust long-term growth for commercial aviation. A key aspect of this is fuel consumption and carbon emissions. With open-source flight data increasingly available, a capability to model fuel burn associated with real-world operations is desired. Existing methods rely on physics-based aircraft performance models and are therefore difficult to scale up due to computation costs. In this study, a data-based approach is explored wherein machine learning techniques are utilized to develop a prediction model for fuel burn, given real-world flight data. Two Machine Learning models, the decision tree and random forest, are trained using operational, weight, and en-route weather variables. The comparisons indicate that Random Forest consistently outperformed Decision Tree for each condition. Takeoff weight is an important predictor, and using only mean values from weather observations was sufficient. Overall, our results confirm that machine learning is a scalable tool to estimate fuel burn and the potential for more extensive integration of environmental metrics into analyses of the aviation industry.