SpaceX Falcon 9 First Stage Landing Prediction Project скачать в хорошем качестве

SpaceX Falcon 9 First Stage Landing Prediction Project

Every successful Falcon 9 landing isn’t just a cool video. It is roughly $62M saved. Every failure is a booster at the bottom of the ocean. That question kept bothering me, so I turned it into a full end‑to‑end project: Can we predict Falcon 9 first-stage landings, and learn why they succeed or fail? In the demo video attached, I walk through what I built and what I learned: I built a multi‑source data pipeline from the official SpaceX REST API + Wikipedia launch tables, cleaned 17 features across 90+ launches, and engineered temporal features like booster age, rolling success rates, and flight sequence. I trained 6 different models (XGBoost, LightGBM, CatBoost, stacking ensembles, etc.), and the surprise winner was a Support Vector Machine with RBF kernel, hitting 94.4% accuracy, outperforming the more complex ensemble methods on this small, structured dataset. The real value is the insights: Boosters with ≥5 flights show 100% landing success in the data. Experience compounds. LEO missions land at ~78%, GTO at ~52% – same rocket, same teams, different energy budget. Orbit profile directly trades off with landing margins. A visible “dip” in success around 2017–2018 wasn’t an operational failure. It lined up with the Block 5 hardware transition. Short‑term regression for long‑term reliability. Using SHAP for interpretability, the model surfaced the same factors SpaceX engineers obsess over: Landing pad used (0.681 importance) Reused count (0.657) Plus grid fins, legs, payload mass, and orbit type. Built a Flask + Gunicorn backend that serves a live landing prediction API. Designed a Three.js + GSAP front‑end with particle rockets, scroll‑based storytelling, interactive charts (Plotly), and an input form where you can plug in launch parameters and see predicted landing probability in real time. Deployed the full stack and turned the analysis into a production web app that anyone can explore. API integration → SQL analysis → feature engineering → model selection → SHAP interpretability → geospatial analysis with Folium → production web app. The biggest lesson for me as a student: Reusable rockets are an engineering miracle, but they’re also a data problem, and data problems can be modeled, explained, and optimized. And if by any chance @Elon Musk or the SpaceX team comes across this: I would love nothing more than to apply this kind of thinking on real missions. 👉 Live app: https://spacex-landing-predictor.onrender.... Let’s connect. I am always open to opportunities (internship, research, or projects)