This project uses machine learning to predict whether a SpaceX Falcon 9 first stage will successfully land and be reused. Successful recovery dramatically reduces launch cost (approx. $165M down to $62M), so forecasting landing outcomes provides financial insight for satellite operators planning missions.
- Collect and prepare Falcon 9 launch data from SpaceX API
- Engineer features related to flight history, payload, site, and booster config
- Train and tune ML models to classify landing success
- Estimate expected launch cost based on predicted reuse probability
- Provide risk guidance for mission planning
- Data: SpaceX v4 API (90 Falcon 9 launches through 2020)
- Models: Logistic Regression, SVM, Decision Tree, KNN
- Techniques: One-hot encoding, standardization, GridSearchCV, confusion matrices
- Metrics: Cross-validation accuracy, test accuracy
- Best models: Logistic Regression & SVM (~0.83 test accuracy)
- Demonstrated cost-impact calculation using landing probability
- Example inference for a new mission:
- Landing success probability: ~85%
- Estimated launch cost: ~$77M
- Risk level: Low
- Booster landing probability drives launch-cost expectation
- Customers can use probability-based pricing logic to:
- Negotiate contracts when risk is higher
- Compare SpaceX vs. traditional launch pricing
- Allocate insurance and contingency spending
- More historical data will improve model robustness, especially across orbit classes and booster configurations
- Incorporate telemetry & weather data
- Deploy as interactive dashboard
- Add calibration + ROC-AUC reporting