Data and code for "Explainable Surrogate Models for Target-Driven Early Design of Ice Arenas"

The package includes: the 2,995 simulated designs (design variables X and decision objectives Y) with their training/test split, and the five runs without valid output; the fixed operating parameters with their sources; the envelope options and unit costs; the trained XGBoost surrogates, with tuned hyperparameters and the accuracy of all four model types; SHAP values for all designs, global importance, interactions, non-linear breakpoints and cost slopes; the decision ranges (maximum admissible clear height and envelope bounds); 28 additional validation simulations; a README with the variable dictionary and full references for all parameter sources, and a Python script that reproduces every result from the data and models.