AirfoilAD: A Benchmark Aerodynamic Dataset for Machine Learning–Based Modeling of Lift and Drag Coefficients

Accurate prediction of aerodynamic coefficients is vital for enhancing airfoil performance in applications ranging from aviation and unmanned aerial vehicles to renewable energy systems. However, the scarcity of publicly available datasets that cover diverse operating conditions, particularly at low Reynolds numbers, limits the effectiveness and generalisation of Machine Learning (ML) models in aerodynamic analysis. This study presents AirfoilAD, a comprehensive dataset that combines experimental and Computational Fluid Dynamics (CFD) data across a wide range of Angles of Attack (AoA), flow velocities, and oscillation frequencies. Using AirfoilAD, two ML algorithms, Random Forest and XGBoost, are developed to predict the lift coefficient (CL[jls-end-space/]) and drag coefficient (CD[jls-end-space/]). XGBoost achieves superior performance, with mean absolute errors below 1% and overall prediction accuracy exceeding 98%, while Random Forest delivers comparably strong yet slightly less accurate results. Feature importance analysis reveals AoA as the dominant factor influencing CL[jls-end-space/], and velocity as the primary driver for CD[jls-end-space/]. By providing a robust and versatile dataset alongside a performance benchmark of advanced ML models, this work bridges a critical gap in aerodynamic resources and establishes a foundation for future developments in real-time flow prediction, control, and inverse airfoil design.