SpaceX Falcon 9 First Stage Landing Prediction

Author: Anik Tahabilder Department: Computer Science, Wayne State University Course: IBM Data Science Professional Certificate - Applied Data Science Capstone

Executive Summary

This project predicts whether SpaceX's Falcon 9 first stage rocket will successfully land after launch. SpaceX advertises Falcon 9 rocket launches at $62 million, significantly cheaper than competitors charging $165+ million per launch. This cost advantage stems from SpaceX's ability to reuse the first stage booster. By predicting landing success, we can estimate launch costs and provide valuable insights for companies bidding against SpaceX for rocket launch contracts.

Key Result: The Decision Tree Classifier achieved the best cross-validation accuracy of 87.5% with test accuracy reaching 88.89-94.44%, making it the recommended model for predicting Falcon 9 first stage landing outcomes.

Business Problem

Problem Statement

SpaceX's revolutionary approach to rocket reusability has disrupted the space launch industry. If we can accurately predict whether the Falcon 9 first stage will land successfully, we can:

Estimate launch costs - A successful landing means the booster can be reused, reducing costs
Support competitive bidding - Alternative companies can make informed bids against SpaceX
Understand success factors - Identify which variables most influence landing success

Why This Matters

Cost Difference: ~$103 million savings per launch when first stage is recovered
Industry Impact: Enables competitors to better understand SpaceX's cost structure
Data-Driven Decisions: Provides quantifiable predictions for launch cost estimation

Dataset Overview

Data Sources

SpaceX REST API - Primary source for launch data
Wikipedia Web Scraping - Supplementary launch records

Dataset Characteristics

Attribute	Description
Records	90 Falcon 9 launches (filtered from 94 total)
Date Range	June 2010 - November 2020
Target Variable	`Class` (1 = Successful Landing, 0 = Failed Landing)
Features	83 features after one-hot encoding

Key Features

Feature	Type	Description
`FlightNumber`	Numeric	Sequential launch number
`PayloadMass`	Numeric	Mass of payload in kg (Mean: ~6,104 kg)
`Orbit`	Categorical	Target orbit (LEO, GTO, ISS, PO, etc.)
`LaunchSite`	Categorical	Launch location (CCAFS SLC 40, KSC LC 39A, VAFB SLC 4E)
`Flights`	Numeric	Number of flights for the booster
`GridFins`	Boolean	Whether grid fins were used
`Reused`	Boolean	Whether the booster was previously used
`Legs`	Boolean	Whether landing legs were deployed
`Block`	Numeric	Booster block version (1-5)
`ReusedCount`	Numeric	Number of times booster was reused

Data Preprocessing

Missing Values: PayloadMass nulls (5 records) imputed with mean value
Encoding: One-hot encoding applied to categorical variables (Orbit, LaunchSite, LandingPad, Serial)
Standardization: StandardScaler applied for ML model training
Train/Test Split: 80/20 split with random_state=2

Methodology

This project follows the CRISP-DM (Cross-Industry Standard Process for Data Mining) methodology:

1. Business Understanding    -> Define prediction goal and success criteria
2. Data Understanding        -> Collect and explore SpaceX launch data
3. Data Preparation          -> Clean, transform, and engineer features
4. Modeling                  -> Train and tune multiple ML algorithms
5. Evaluation                -> Compare models using cross-validation
6. Deployment                -> Document findings and recommendations

Tools and Technologies

Category	Tools
Programming	Python 3.x
Data Manipulation	Pandas, NumPy
Visualization	Matplotlib, Seaborn, Folium
Machine Learning	Scikit-learn
Database	SQLite, SQL
Web Scraping	BeautifulSoup, Requests

Data Collection

1. SpaceX API Data Collection

Notebook: jupyter-labs-spacex-data-collection-api.ipynb

# API Endpoint
spacex_url = "https://api.spacexdata.com/v4/launches/past"

# Data extraction workflow:
# 1. GET request to SpaceX API
# 2. Parse JSON response with pd.json_normalize()
# 3. Extract booster version, launch site, payload, and core data
# 4. Filter for Falcon 9 launches only
# 5. Handle missing values

Key API Endpoints Used:

/v4/launches/past - Historical launch data
/v4/rockets/{id} - Booster version details
/v4/launchpads/{id} - Launch site coordinates
/v4/payloads/{id} - Payload mass and orbit
/v4/cores/{id} - Core reuse information

2. Web Scraping Wikipedia

Notebook: jupyter-labs-webscraping.ipynb

# Target URL (snapshot from June 9, 2021)
static_url = "https://en.wikipedia.org/w/index.php?title=List_of_Falcon_9_and_Falcon_Heavy_launches&oldid=1027686922"

# Extraction process:
# 1. HTTP GET request with custom headers
# 2. Parse HTML with BeautifulSoup
# 3. Extract table data from <th> and <td> elements
# 4. Clean and normalize data

Scraped Data: 121 launch records with flight details, outcomes, and booster landing status.

Exploratory Data Analysis

SQL Analysis

Notebook: jupyter-labs-eda-sql-coursera_sqllite.ipynb

Key SQL queries and findings:

-- Unique launch sites
SELECT DISTINCT Launch_Site FROM SPACEXTABLE;
-- Results: CCAFS LC-40, VAFB SLC-4E, KSC LC-39A, CCAFS SLC-40

-- Total payload mass for NASA (CRS) missions
SELECT SUM(PAYLOAD_MASS__KG_) FROM SPACEXTABLE WHERE Customer = 'NASA (CRS)';
-- Result: 45,596 kg

-- Average payload for F9 v1.1
SELECT AVG(PAYLOAD_MASS__KG_) FROM SPACEXTABLE WHERE Booster_Version = 'F9 v1.1';
-- Result: 2,928.4 kg

-- First successful ground pad landing
SELECT MIN(Date) FROM SPACEXTABLE WHERE Landing_Outcome = 'Success (ground pad)';
-- Result: 2015-12-22

Visualization Analysis

Notebook: jupyter-labs-eda-dataviz-v2.ipynb

Key Visualizations Created:

Flight Number vs. Launch Site - Launch frequency patterns
Payload Mass vs. Launch Site - Payload capacity by location
Success Rate by Orbit Type - Orbit-specific success rates
Yearly Success Trend - Improvement over time

Key Insights:

Success rate increased significantly after 2015
KSC LC-39A has the highest success rate among launch sites
LEO and ISS orbits show strong correlation between flight number and success
Heavy payloads (>10,000 kg) are not launched from VAFB-SLC

Interactive Visual Analytics

Folium Maps Analysis

Notebook: lab-jupyter-launch-site-location-v2.ipynb

Interactive maps were created to analyze:

Launch Site Locations
- All sites are in proximity to the equator (favorable for orbital mechanics)
- All sites are close to coastlines (safety for failed launches)
Success/Failure Markers
- Color-coded markers (green=success, red=failure)
- Cluster visualization for overlapping launches
Proximity Analysis
- Distance to coastline
- Distance to railways, highways, and cities
- Launch sites maintain safe distances from populated areas

Launch Sites Analyzed:

Site	Location	Coordinates
CCAFS LC-40	Cape Canaveral, FL	28.56°N, 80.58°W
CCAFS SLC-40	Cape Canaveral, FL	28.56°N, 80.58°W
KSC LC-39A	Kennedy Space Center, FL	28.61°N, 80.60°W
VAFB SLC-4E	Vandenberg AFB, CA	34.63°N, 120.61°W

Machine Learning Prediction

Notebook: SpaceX-Machine-Learning-Prediction-Part-5-v1.ipynb

Data Preparation

# Feature matrix: 83 features after one-hot encoding
X = pd.read_csv('dataset_part_3.csv')

# Target variable
Y = data['Class'].to_numpy()

# Standardization
transform = preprocessing.StandardScaler()
X = transform.fit_transform(X)

# Train-Test Split
X_train, X_test, Y_train, Y_test = train_test_split(X, Y, test_size=0.2, random_state=2)
# Training samples: 72, Test samples: 18

Models Evaluated

Four classification algorithms were evaluated using GridSearchCV with 10-fold cross-validation:

1. Logistic Regression

parameters = {
    'C': [0.01, 0.1, 1],
    'penalty': ['l2'],
    'solver': ['lbfgs']
}

Best Parameters: C=0.01, penalty='l2', solver='lbfgs' Cross-Validation Accuracy: 84.64% Test Accuracy: 83.33%

2. Support Vector Machine (SVM)

parameters = {
    'kernel': ('linear', 'rbf', 'poly', 'sigmoid'),
    'C': np.logspace(-3, 3, 5),
    'gamma': np.logspace(-3, 3, 5)
}

Best Parameters: C=1.0, gamma=0.0316, kernel='sigmoid' Cross-Validation Accuracy: 84.82% Test Accuracy: 83.33%

3. Decision Tree Classifier

parameters = {
    'criterion': ['gini', 'entropy'],
    'splitter': ['best', 'random'],
    'max_depth': [2, 4, 6, 8, 10, 12, 14, 16, 18],
    'max_features': ['sqrt', 'log2', None],
    'min_samples_leaf': [1, 2, 4],
    'min_samples_split': [2, 5, 10]
}

Best Parameters:

criterion='entropy'
max_depth=6
max_features='log2'
min_samples_leaf=1
min_samples_split=5
splitter='best'

Cross-Validation Accuracy: 87.5% Test Accuracy: 88.89-94.44%

4. K-Nearest Neighbors (KNN)

parameters = {
    'n_neighbors': [1, 2, 3, 4, 5, 6, 7, 8, 9, 10],
    'algorithm': ['auto', 'ball_tree', 'kd_tree', 'brute'],
    'p': [1, 2]
}

Best Parameters: algorithm='auto', n_neighbors=6, p=1 Cross-Validation Accuracy: 84.46% Test Accuracy: 94.44%

Model Comparison and Selection

Performance Summary

Model	CV Accuracy	Test Accuracy	Precision	Recall	F1-Score
Decision Tree	87.50%	88.89%	1.000	0.857	0.923
SVM	84.82%	83.33%	0.929	0.929	0.929
Logistic Regression	84.64%	83.33%	0.933	1.000	0.966
KNN	84.46%	94.44%	0.933	1.000	0.966

Why Decision Tree Was Selected

The Decision Tree Classifier was selected as the best model for the following reasons:

1. Highest Cross-Validation Accuracy (87.5%)

Most reliable indicator of generalization performance
Less susceptible to overfitting on small test sets

2. Interpretability

Decision rules can be visualized and understood
Important for stakeholders who need to explain predictions
Clear feature importance rankings

3. No Feature Scaling Required

Works with both numerical and categorical features natively
Less preprocessing complexity

4. Handles Non-Linear Relationships

Captures complex interactions between features
Orbit type, payload mass, and booster version interact non-linearly

Why Not Other Models?

Model	Reason Not Selected
Logistic Regression	Lower CV accuracy (84.64%); assumes linear relationships which may not hold for rocket landing physics
SVM	Similar CV accuracy but less interpretable; sigmoid kernel may not capture all patterns
KNN	High test accuracy but lower CV accuracy (84.46%); computationally expensive at prediction time; sensitive to feature scaling

Confusion Matrix Analysis (Decision Tree)

                 Predicted
              |  No Land  |  Landed  |
Actual -------|-----------|----------|
  No Land     |     3     |    0     |  (True Negatives, False Positives)
  Landed      |     2     |   13     |  (False Negatives, True Positives)

Strength: Zero false positives (never incorrectly predicts success)
Area for Improvement: Some false negatives (missed successful landings)

Key Findings

Technical Findings

Flight Experience Matters
- Success rate increases with flight number
- Later launches (post-2015) show significantly higher success rates
Launch Site Influence
- KSC LC-39A has the highest success rate
- All sites are strategically located near coastlines
Payload Mass Impact
- Moderate payloads (2,000-6,000 kg) show good success rates
- Very heavy payloads have mixed results depending on orbit
Booster Reuse Success
- Reused boosters with grid fins and legs show high success rates
- Block 5 boosters demonstrate the best performance

Business Implications

Cost Estimation: Successful landing prediction enables accurate cost modeling
Risk Assessment: Identify high-risk launch configurations
Competitive Intelligence: Understand SpaceX's operational patterns

Conclusion

This capstone project successfully demonstrates the application of data science methodologies to a real-world aerospace problem. By collecting data from multiple sources, performing thorough exploratory analysis, and training multiple machine learning models, we achieved:

Prediction Accuracy: 87.5% cross-validation accuracy with Decision Tree
Actionable Insights: Identified key factors influencing landing success
Business Value: Enabled launch cost estimation for competitive bidding

The Decision Tree model provides the best balance of accuracy and interpretability for predicting Falcon 9 first stage landing success.

How to Run This Project

Prerequisites

pip install pandas numpy matplotlib seaborn scikit-learn folium beautifulsoup4 requests sqlalchemy

Execution Order

Run the notebooks in the following sequence:

Data Collection

jupyter-labs-spacex-data-collection-api.ipynb
jupyter-labs-webscraping.ipynb

Exploratory Data Analysis

jupyter-labs-eda-sql-coursera_sqllite.ipynb
jupyter-labs-eda-dataviz-v2.ipynb

Interactive Analytics

lab-jupyter-launch-site-location-v2.ipynb

Machine Learning

SpaceX-Machine-Learning-Prediction-Part-5-v1.ipynb

Project Structure

IBM-Data-Science-Certificate/
|
|-- jupyter-labs-spacex-data-collection-api.ipynb  # API data collection
|-- jupyter-labs-webscraping.ipynb                  # Web scraping from Wikipedia
|-- jupyter-labs-eda-sql-coursera_sqllite.ipynb    # SQL-based analysis
|-- jupyter-labs-eda-dataviz-v2.ipynb              # EDA with visualizations
|-- lab-jupyter-launch-site-location-v2.ipynb      # Folium interactive maps
|-- SpaceX-Machine-Learning-Prediction-Part-5-v1.ipynb  # ML models
|
|-- dataset/                                        # Data files directory
|-- figures/                                        # Generated visualizations
|-- README.md                                       # This file

Acknowledgments

IBM Skills Network - Course materials and lab infrastructure
SpaceX - Public API providing launch data
Wikipedia - Historical launch records

Contact

Anik Tahabilder Department of Computer Science Wayne State University

This project was completed as part of the IBM Data Science Professional Certificate program.

Name		Name	Last commit message	Last commit date
Latest commit History 14 Commits
.ipynb_checkpoints		.ipynb_checkpoints
README.md		README.md
SpaceX-Machine-Learning-Prediction-Part-5-v1.ipynb		SpaceX-Machine-Learning-Prediction-Part-5-v1.ipynb
jupyter-labs-eda-dataviz-v2.ipynb		jupyter-labs-eda-dataviz-v2.ipynb
jupyter-labs-eda-sql-coursera_sqllite.ipynb		jupyter-labs-eda-sql-coursera_sqllite.ipynb
jupyter-labs-spacex-data-collection-api.ipynb		jupyter-labs-spacex-data-collection-api.ipynb
jupyter-labs-webscraping.ipynb		jupyter-labs-webscraping.ipynb
lab-jupyter-launch-site-location-v2.ipynb		lab-jupyter-launch-site-location-v2.ipynb

Folders and files

Latest commit

History

Repository files navigation