Data-driven prediction of runway incursions using random forest and temporal validation

This study develops a predictive framework for runway incursions at major U.S. airports by integrating aviation-specific feature engineering with Random Forests and temporal validation. Using data from 60 hub airports (2015–2024), composite indices captured weather, infrastructure, and operational pressures. Random Forest outperformed mean, OLS, and Poisson baselines (R²=0.789, RMSE=2.99). Cross-validation confirmed stable generalization (R²=0.757, 95% CI [0.629, 0.885]). Statistical tests showed significant COVID-19 decline and identified VFR conditions as the strongest predictor. Outliers were retained for robustness, and multicollinearity checks validated Random Forest resilience. The framework complements the European Organization for the Safety of Air Navigation (EUROCONTROL) guidance, supporting risk-based staffing, monitoring, and training.