This project is an enhanced adaptation of my IBM Data Science Capstone project, delivering an end-to-end data science solution to predict the landing success of the SpaceX Falcon 9 rocket's first stage. The original capstone project was completed as part of my IBM Data Science Professional Certificate on Coursera.

The Falcon 9 is a revolutionary vehicle in the aerospace industry due to its reusable first stage, which dramatically reduces launch costs. The ability to accurately predict landing success is critical for cost estimation, operational planning, and competitive analysis in the commercial launch market.

This project demonstrates a complete workflow from data acquisition to predictive modeling, providing actionable insights into the factors driving launch success.

The primary goals of this project were to:

Identify the key factors—such as payload mass, booster version, orbit type, and launch site—that influence Falcon 9 first-stage landing success.

Quantify performance differences between launch sites and booster versions.

Explore how payload ranges and mission profiles affect landing outcomes.

Build and evaluate a predictive model capable of accurately forecasting landing success to support decision-making.

The project followed a comprehensive data science pipeline:

Data Collection: Combined data from the SpaceX REST API and web-scraped Wikipedia to build comprehensive datasets of Falcon 9 launches.

Data Wrangling: Cleaned, standardized, and engineered key features, including payload mass, booster version, orbit type, and the target landing outcome variable.

Interactive Visualization: Built an interactive Plotly Dash dashboard for dynamic exploration of launch data and success rates.

Predictive Modeling: Trained, tuned, and evaluated multiple classification models, including Logistic Regression, Support Vector Machine (SVM), Decision Tree, and K-Nearest Neighbors (KNN), using GridSearchCV for hyperparameter optimization.

After training and evaluation, the performance of the classification models was compared based on accuracy.

Model Training Accuracy Test Accuracy Decision Tree 88.9% 94.4% Support Vector Machine (SVM) 86.2% 83.6% Logistic Regression 83.4% 83.3% K-Nearest Neighbors (KNN) 83.4% 77.8%

The Decision Tree classifier was the best-performing model with a 94.4% test accuracy, demonstrating excellent predictive power for landing success.

Logistic Regression and SVM showed strong and consistent performance, making them reliable candidates for scenarios where model interpretability is key.

Data analysis revealed that booster upgrades and payload optimization are critical factors in improving landing reliability. Furthermore, launch site KSC LC-39A demonstrated the highest success rate.

Some notebooks include interactive Plotly and Folium visualizations that may not render correctly on GitHub. To view the fully interactive version, launch it on Binder:

Ensure you have Python 3.7 or later installed on your machine.

Clone the repository:

Install required libraries:

Run the Jupyter Notebooks: Execute the notebooks in numerical order (from 1- to 8-) to follow the project's workflow from data collection to prediction.

Launch the Dash App (Optional):

Languages: Python, SQL

Data Handling: Pandas, NumPy

Data Collection: Requests, Beautiful Soup

Database & Analytics: SQLAlchemy, ipython-sql

Visualization: Matplotlib, Seaborn, Plotly, Folium

Machine Learning: Scikit-learn

Web Application: Dash

Filipe Braiman Carvalho Applied AI & LLM Systems | Deep Learning · Transformers · RAG · Computer Vision | End-to-End ML Engineering

Email: filipebraiman@gmail.com LinkedIn: linkedin.com/in/filipe-b-carvalho GitHub: github.com/filipe-braiman

AI and data professional with experience in LLM evaluation, retrieval-augmented generation (RAG), and AI model validation. Currently working in AI R&D at Huawei as an AI Evaluation Specialist, contributing to the reliability and real-world performance of LLM and RAG systems. Strong background in Python-based data and AI workflows, including model assessment, dataset development, and analytical reporting for production-oriented AI solutions. Portfolio projects explore deep learning architectures, computer vision, RAG systems, and applied machine learning experimentation, emphasizing reproducible ML pipelines and practical AI engineering.

Version Date Changes 1.0 2025-10-29 First publication. 1.1 2026-03-08 Updated Author section.