An interpretable hybrid learning model for predicting Falcon 9 booster recovery success

Abstract The challenge of reliably recovering reusable launch vehicle boosters defines the future of affordable and sustainable space transportation. Addressing this challenge, we developed an interpretable hybrid machine learning model to predict the recovery success of SpaceX Falcon 9 first-stage boosters. Our model synergizes an SVM with an ANN and leverages MShOA for simultaneous feature selection and hyperparameter tuning, thereby enabling robustness and predictive performance even with small, imbalanced datasets. Evaluated on the public Kaggle Falcon 9 dataset using stratified cross-validation, the model achieved an AUC of 0.956 and an F1-score of 0.896. SHAP-based analysis identified ReusedCount, Legs, and Flights as key determinants. These results show that interpretable hybrid models yield reliable, data-driven insights for reusable launch vehicle operations and support mission planning and risk assessment.