Analysis of Factors affecting Injury Severity in Aircraft Accidents based on Machine Learning

This study utilized accident data from the National Transportation Safety Board (NTSB) to develop a predictive model for factors affecting human injury severity in aircraft accidents using the XGBoost algorithm. After addressing class imbalance, the model achieved accuracy of 0.850, recall of 0.842, and F1 score of 0.815 at a learning rate of 0.7. SHAP analysis identified distance to the airport, elapsed time since inspection, cumulative flight hours, seasonal factors, and wind speed as the most influential factors. Unlike previous studies that primarily focused on mechanical causes, this study demonstrates the importance of environmental and operational conditions, highlighting the need for data-driven safety management and practical implications for real-time risk detection and improving global ATC operations.