Falcon 9 Landing Prediction - IBM Data Science Capstone Project

Welcome to the Falcon 9 Landing Prediction repository, part of the IBM Data Science Professional Certificate. This project leverages machine learning to predict the reusability of the Falcon 9 rocket's first stage, aiming to reduce the cost of space launches by forecasting landing success.

Table of Contents

1. Project Overview
2. Motivation
3. Data Sources
4. Technologies Used
5. Setup Instructions
6. Modeling Process
   • Data Preprocessing
   • Model Training
7. Exploratory Data Analysis
8. Results
9. Dashboard Implementation
   • Dashboard Features
   • Interactive Plots
10. Future Work
11. Appendix
12. Contributors
13. License

---

Project Overview

This project simulates a scenario where Space Y, a new competitor to SpaceX, uses machine learning techniques to analyze the reusability of the Falcon 9 first stage. By predicting successful landings, Space Y aims to optimize launch costs and increase the sustainability of its space program.

---

Motivation

The reusability of rockets is crucial in reducing the costs of space missions. By understanding when the Falcon 9 first stage successfully lands, this project can provide insights that help optimize operational decisions and cost structures for future launches. Machine learning allows us to analyze historical launch data and forecast the landing outcomes.

---

Data Sources

The dataset used in this project includes public data from SpaceX, which was obtained from:

• SpaceX API: For fetching launch data and outcomes.
• Wikipedia: Web scraping using BeautifulSoup to collect historical data on Falcon 9 and Falcon Heavy launches.
• Additional Data: Launch dates, payload mass, orbital parameters, launch sites, and landing outcomes covering launches from 2010 to 2020.

---

Technologies Used

This project utilizes the following technologies and libraries:

• Python: Main programming language.
• Pandas: Data analysis and manipulation.
• NumPy: Numerical computations.
• Scikit-learn: Machine learning library for model development.
• Matplotlib & Seaborn: For visualizing data trends and patterns.
• Plotly Dash: Used to build interactive dashboards.
• Folium: To create interactive maps for visualizing launch sites.
• Jupyter Notebook: For interactive analysis and development.
• SQLite & SQLAlchemy: For SQL queries and data management.

---

Setup Instructions

1. Clone the repository:

git clone https://github.com/yourusername/falcon9-landing-prediction-ibm-capstone.git
cd falcon9-landing-prediction-ibm-capstone

2. Install the required dependencies:

pip install -r requirements.txt

3. Launch Jupyter Notebooks for exploration and model training:

jupyter notebook

4. Run the Plotly Dash dashboard:

python spacex_dash_app.py

---

Modeling Process

Data Preprocessing

• Data Cleaning: Missing values were imputed or removed to maintain dataset integrity.
• Feature Engineering: Used get_dummies() for one-hot encoding of categorical variables like Orbit, LaunchSite, and LandingPad.
• Scaling: All numeric columns were standardized using StandardScaler() to optimize model performance.

Model Training

Several machine learning models were trained, tuned, and evaluated:

• Logistic Regression: Basic classifier used as a baseline.
• Support Vector Machine (SVM): Tested with various kernels for boundary classification.
• Decision Trees: Used for rule-based classification of landing outcomes.
• K-Nearest Neighbors (KNN): Applied to cluster data based on proximity.
• GridSearchCV: Hyperparameter tuning to optimize the model.

---

Exploratory Data Analysis

Key visualizations produced during the EDA phase included:

• Flight Number vs. Launch Site: Analyzed how launch success varied across different sites.
• Payload vs. Orbit: Identified the relationship between payload mass and orbital success.
• Launch Success Yearly Trend: Visualized trends in landing success over time.
• Success Rate by Orbit Type: Compared landing success rates across different orbits.
• Total Payload by NASA: Showcased the total payload mass for NASA-sponsored launches.

---

Results

The following accuracies were achieved on the test data:

• Logistic Regression: 83.3%
• SVM: 83.3%
• Decision Trees: 88.9% (Best Performing)
• K-Nearest Neighbors (KNN): 83.3%

Confusion Matrix for Best Model:

The Decision Tree model achieved the highest accuracy with an accuracy score of 88.9%. The confusion matrix showed a relatively high true positive rate, but a small percentage of false positives.

---

Dashboard Implementation

Dashboard Features

• Launch Site Selection: Users can filter by launch site to view success/failure ratios.
• Payload Range Slider: Interactive slider to filter launches by payload mass.
• Pie Charts: Visual representation of success/failure ratios per launch site.
• Scatter Plots: Payload vs. Launch Outcome scatter plots color-coded by booster version.

Interactive Plots

• Success Pie Charts: Showed the success rate for all launch sites.
• Payload vs. Outcome Scatter Plot: Helped identify trends in payload mass and landing success.

---

Future Work

• Incorporate Weather Data: Introduce weather data to refine the landing prediction model.
• Advanced Modeling: Test advanced models like Random Forest and XGBoost to improve classification accuracy.
• Deployment: Deploy the prediction model using Flask or FastAPI for real-time predictions.

---

Appendix

Relevant assets for this project include:

• SQL Queries: Used to filter and aggregate data (e.g., payload mass, landing outcomes).
• Code Snippets: Python functions for feature engineering and model evaluation.
• Charts: Visualizations produced during EDA (e.g., bar charts, scatter plots).
• Notebook Outputs: Key results from Jupyter notebooks used in the project.

---

Contributors

• Manuel Luján Vilchez - Data Scientist and Project Lead
• IBM Data Science Capstone Project Team - Instructors and resources for guidance.

---

License

This project is licensed under the MIT License. See the LICENSE file for more information.