🚀 SpaceX Launch Success Prediction with Machine Learning

How I predicted SpaceX rocket landing success using real launch data and interactive dashboards.

Role: Data Scientist

Team Size: Individual project

Duration: 3 weeks

Stack: Python · Pandas · SQL · Plotly Dash · Folium · REST APIs · Scikit-learn · SVM · Decision Trees · Logistic Regression

🧠 Project Overview

This project was developed as the final capstone for the IBM Data Science Professional Certificate. The objective was to analyze SpaceX's Falcon 9 launch data to uncover which factors influence the success of first-stage rocket landings and to build machine learning models that predict landing outcomes.

By integrating data science techniques — from API usage and web scraping to EDA, dashboarding, and machine learning — this project demonstrates an end-to-end approach to solving a real-world aerospace problem using data.

🔍 Problem Statement

SpaceX can significantly reduce launch costs by recovering and reusing the first stage of its Falcon 9 rockets. However, not all launches result in successful landings. Key questions addressed:

• Can we predict whether a rocket will land successfully based on launch characteristics?

• Which features (payload, orbit, launch site, etc.) are most influential?

📊 Data Collection & Processing

Sources:

• SpaceX REST API

• Wikipedia Falcon 9 Launch History

Steps:

• Data retrieved using Python requests and web scraping (BeautifulSoup).

• Cleaned and filtered for Falcon 9 launches.

• Missing values were handled (e.g., payload mass imputation).

• Binary target variable created: landing_class (1 = success, 0 = fail).

📈 Exploratory Data Analysis (EDA)

Techniques used:

• SQL queries on launch data for patterns (e.g., payload averages, site success rates).

• Visual analysis with seaborn, Matplotlib, and pandas.

• Key findings:

  • Launch site, payload mass, and booster version have strong correlation with landing outcomes.

  • Sites like KSC LC-39A had the highest success rate (~77%).

  • Best results were achieved with payloads between 2000–5500kg and booster version FT.

🌍 Interactive Visualization

Tools: Folium, Plotly Dash

1. Folium Map:

   • Visualized launch site locations.

   • Overlayed launch outcomes and distances to coastlines and airports.

2. Dashboard (Plotly Dash):

   • Pie Chart: Displays success rate per launch site.

   • Payload Slider + Scatter Plot: Shows success rate correlation with payload and booster version.

   • Fully interactive — adjust site and payload range to explore outcomes.

   
   

🤖 Machine Learning Modeling

Goal: Predict success of Falcon 9 first-stage landings.

Models trained:

• Logistic Regression

• Support Vector Machine (SVM)

• Decision Tree

• K-Nearest Neighbors (KNN)

Approach:

• Scaled features, split dataset (80/20).

• Hyperparameter tuning for all models.

• Evaluation via accuracy score and confusion matrix.

Results:

• All models achieved ~83% test accuracy.

• False positives (predicting success when it failed) were common, due to small dataset.

📌 Key Takeaways

• It’s feasible to build a predictive model for SpaceX landing success using historical data.

• Features like launch site, payload mass, and booster version significantly impact the outcome.

• Interactive dashboards and geospatial visualization enhance exploratory capabilities.

• Data limitations (size and imbalance) affect generalization — future work could include more recent missions or additional parameters (e.g., weather, telemetry).

🔗 Project Assets

• 📂 GitHub Repository: Capstone SpaceX Launch Success

• 📍 Presentation Slides: See full breakdown of analysis here