Fuel Burn Prediction from Open Flight Trajectory and Meteorological Data for the PRC Data Challenge 2025

This repository contains the methodology and solutions developed for the Performance Review Commission (PRC) Data Challenge 2025. The main objective of this work is to provide a comprehensive, data-driven approach for predicting aircraft fuel burn by integrating open flight trajectories, meteorological reanalysis, and advanced machine learning techniques. The project leverages a hybrid approach that combines physical modeling (mass estimation, flight phase detection) with ensemble learning models (XGBoost, LightGBM) to achieve high-fidelity predictions. The following artefacts are present in this repository:Source Code: The complete Python-based pipeline for scalable data preprocessing (using duckdb and pandas), trajectory mining, and meteorological augmentation (ERA5 and METAR/ASOS). Predictive Models: Implementation of optimized gradient boosting regressors with hyperparameter tuning performed via the Optuna framework. Manuscript: A research-oriented paper detailing the feature engineering process, including the derivation of physically motivated variables like true airspeed (TAS), aircraft mass estimation using the total energy model, and phase-specific fuel-flow summaries. Complementary Data: Integration of publicly available external sources, including the FAA Aircraft Characteristics Database and OpenAP performance models, ensuring reproducibility and open-science compliance.