Can we predict whether a Falcon 9 first stage will land? SpaceX lists launches at about $62 million, while other providers charge upward of $165 million, largely because SpaceX reuses the first stage. Predicting a successful landing is a way to estimate the cost of a launch.
Capstone project for the IBM Data Science Professional Certificate (2022): the full data science workflow, from collecting data through an API and web scraping to a dashboard and a machine learning model.
- 90 launches in the final dataset; 67% of first stages landed successfully.
- Landing success climbed from 0% in 2010–2013 to 90% in 2019.
- Launches to SSO, HEO, GEO and ES-L1 orbits always landed; GTO launches landed about half the time.
- Four classifiers, each tuned with 10-fold grid search:
| Model | Cross-validation accuracy | Test accuracy (18 launches) |
|---|---|---|
| Decision tree | 87.3% | 83.3% |
| Support vector machine | 84.8% | 83.3% |
| k-nearest neighbors | 84.8% | 83.3% |
| Logistic regression | 84.6% | 83.3% |
The decision tree scored best in cross-validation; on the small test set all four models tied.
| Step | Notebook / file | What it does |
|---|---|---|
| 1. Collect (API) | Data_Collection_API.ipynb |
Pulls launch data from the SpaceX REST API, keeps Falcon 9 launches, fills missing payload masses |
| 2. Collect (scraping) | Data Collection_WebScraping.ipynb |
Scrapes Falcon 9 launch records from Wikipedia with BeautifulSoup |
| 3. Wrangle | DataWrangling.ipynb |
Turns landing outcomes into a success (1) / failure (0) label |
| 4. Explore with SQL | EDAwithSQL.ipynb |
Queries launch sites, payloads, boosters and outcomes |
| 5. Explore with charts | EDAwithDataViz.ipynb |
Success by flight number, payload, orbit and year; one-hot encodes features |
| 6. Map | VisualAnalytics_Folium.ipynb |
Interactive Folium map of launch sites, outcomes and distances to coast, rail and cities |
| 7. Dashboard | dash_app.py |
Plotly Dash app: success share by site and a payload-range slider |
| 8. Predict | MachineLearningPrediction.ipynb |
Standardizes features, tunes and compares the four classifiers |
The full write-up is in the presentation (PDF).
pip install pandas numpy matplotlib seaborn scikit-learn folium dash requests beautifulsoup4 jupyterjupyter notebookpython dash_app.pyThe dashboard runs at http://127.0.0.1:8050. The CSV files in the repository are the outputs of each step, so any notebook can run on its own.