Explainable Aircraft Crash Severity Prediction Using XGBoost and SHAP

Aircraft crash prediction, as a major research topic, focuses on improving aeronautical safety and minimizing fatalities. This study aims to improve the prediction of accident severity by integrating Explainable AI (XAI) with Machine Learning (ML). We use several different types of ML model, including XGBoost, Decision Tree (DT), Random Forest (RF), and AdaBoost, to model aviation accidents to predict the crash based on a set of characteristics. XGBoost shows higher accuracy than the other models studied. We also employ SHAP (SHapley Additive exPlanations) to enhance explainability by locating the most important features that impact the predictions. To analyze the correlations among the attributes and to detect possible redundancies, a correlation matrix is used. Thus, the computational efficiency of XGBoost is assessed, demonstrating its competency. We cross-validate robustness and perform paired t-tests to confirm the statistical significance of the observed performance variations. Additionally, we validated XGBoost using a second dataset to further confirm its reliability and generalization capability across different data distributions. The results of this study point out the need to combine advanced ML tools with XAI in key safety applications.