Machine learning prediction of aircraft anti-icing fluid endurance: A comparative study using natural and artificial snow data

Ground ice poses a critical safety risk to winter aviation operations, primarily because the protective endurance of aircraft anti-icing fluids fluctuates unpredictably with volatile weather conditions. To mitigate these risks, this study develops a robust machine learning framework designed to forecast fluid endurance times using high-fidelity data from both artificial and natural snow environments. Research was conducted via endurance tests following the SAE ARP5485B protocol to train and evaluate multiple decision tree-based architectures, specifically the Decision Tree, Random Forest, and Extreme Gradient Boosting (XGBoost) models. Model reliability was ensured through rigorous k-fold cross-validation and systematic hyperparameter optimization. Results demonstrate that the optimized XGBoost model achieved superior predictive accuracy, yielding an RMSPE of 15.38% and a MAPE of 8.88%. Furthermore, SHAP analysis identified snow deposition rate and ambient temperature as the most influential determinants of fluid failure. These findings suggest that integrating machine learning into endurance testing can revolutionize current safety protocols and facilitate precise, data-driven decision-making for winter ground operations.