Colloquium: Andrada Ioana Gheorghe (C&O)

Predicting aircraft Take-Off Weight (TOW) has been a highly sought-after task for aviation stakeholders, yet its sensitivity to operational trends and cost indices leads aircraft operators to keep it confidential. This study addresses the need for (pre-)tactical TOW prediction for flight planning and network management applications, solely based on Flight Plan and Terminal Aerodrome Forecast parameters. Several machine learning algorithms were trained with flights at Amsterdam Airport Schiphol (AMS), where Gradient Boosting Decision Trees achieved the smallest Mean Absolute Percentage Error with 1.71% (training) and 2.17% (testing). The most influencing features proved to be requested cruise speed, great circle distance between airports, and aircraft type. Model validation at Paris (CDG) and Charleroi (CRL) demonstrated the model's adaptability at other airports. Future work aims to expand training data to European network level, fostering neural networks implementation. Findings emphasise the importance of considering speed profile features, when aiming for trajectory predictions improvement.

Supervisor: Marta Ribeiro & Junzi Sun