🌌 ExoVision-AI

AI-Powered Exoplanet Detection System | NASA Space Apps Hackathon 2025

“When the stars speak in data, we listen with algorithms.”
Bridging space exploration and machine learning to uncover distant worlds.

---

---

🚀 Overview

ExoVision-AI is a machine learning–driven web application that automatically classifies celestial objects as Confirmed Exoplanets or Planetary Candidates using NASA's open K2 Planets and Candidates Catalog.
Developed by a team of passionate space and AI enthusiasts for the NASA Space Apps Hackathon 2025, this project bridges data science and astrophysics through real-time predictive modeling.

---

🌠 Key Highlights

• Intelligent ML Pipeline – Preprocessing, feature engineering, and optimized Decision Tree + XGBoost models.
• Interactive Web App (Flask) – Upload datasets or manually enter parameters for instant classification.
• Dynamic Visualizations – Distribution charts, correlation heatmaps, ROC curves, and performance metrics.
• End-to-End Reproducibility – Ready-to-run environment with NASA data integration and trained models.
• Hackathon-Optimized – Lightweight, interpretable, and deployable in under 10 minutes.

---

🧬 System Architecture

ExoVision-AI/
│
├── app/
│   ├── static/uploads/         # Uploaded CSVs
│   ├── templates/              # Front-end HTML pages
│   │   ├── index.html
│   │   ├── upload.html
│   │   ├── results.html
│   │   ├── manual_input.html
│   │   ├── manual_results.html
│   │   └── about.html
│   └── app.py                  # Flask web app
│
├── Dataset/
│   └── nasa-archive.csv        # NASA K2 dataset
│
├── Models/
│   ├── decision_tree_model.pkl
│   ├── decision_tree_pipeline.pkl
│   └── xgb_exoplanet_model.pkl
│
├── Notebook/
│   └── nasa-space-app.ipynb    # Full model training & evaluation notebook
│
├── requirements.txt
└── README.md

---

🧩 Machine Learning Pipeline

• Preprocessing

  • Missing-value imputation (SimpleImputer)
  • Feature engineering (stellar & orbital parameters)
  • One-hot encoding for categorical features

• Modeling Strategy

  • Parallel Model Development: Team divided into two sub-teams for comparative analysis
  • Team A: Trained XGBoost and Decision Tree models (selected for final implementation)
  • Team B: Trained LightGBM model (used for performance benchmarking)
  • 5-Fold Stratified Cross-Validation

• Model Selection

  • XGBoost selected for superior performance and interpretability
  • Decision Tree maintained as fallback for transparency and comparison
  • Final ensemble approach leverages both models for robust predictions

• Evaluation Metrics

  • Accuracy, Precision, Recall, F1-Score, ROC-AUC
  • Confusion Matrix & ROC Curve visualizations

Example average CV Accuracy: ~99.0%

---

🖥️ Web Application

Built using Flask, the interface allows two modes:

1. 📤 CSV Upload:
   Upload new K2 data to automatically classify planetary candidates.

2. 🧮 Manual Input:
   Enter stellar and planetary parameters via form input to get instant predictions.

Features include:

• Real-time visualization of dataset distributions
• Heatmaps and ROC curves for interpretability
• Side-by-side comparison of XGBoost vs Decision Tree predictions

---

📸 Suggested Images

• /app/static/architecture.png → Model + Flask integration
• /app/static/app_preview.png → Upload & results interface
• /app/static/visualizations.png → Heatmap / ROC curve examples

![Architecture Overview](app/static/architecture.png)
![App Preview](app/static/app_preview.png)
![Visualizations](app/static/visualizations.png)

---

⚙️ Installation & Setup

# Clone repository
git clone https://github.com/c0llectorr/ExoVision-AI.git
cd ExoVision-AI

# Create virtual environment (recommended)
python -m venv venv
source venv/bin/activate       # Linux/macOS
venv\Scripts\activate        # Windows

# Install dependencies
pip install -r requirements.txt

---

▶️ Run the Application

cd app
python app.py

Then open your browser and navigate to:

http://127.0.0.1:5000/

---

📈 Example Outputs

Metric	Decision Tree	XGBoost
Accuracy	0.99	0.993
Precision	0.98	0.99
Recall	0.99	0.99
F1 Score	0.985	0.992

---

🧑‍🚀 Team ExoVision

Member	Role	Area
Muhammad Ahmad	Team Leader & Data Scientist	Project Leadership, Data Strategy, Model Evaluation
Syed Darain Hyder Kazmi	ML Engineer	Model Training & Deployment
Muhammad Ahsan Atiq	Backend Developer	Flask API Integration
Muhammad Mohsin	Frontend Developer	HTML Templates & UX
Ali Hassan	Research Lead	NASA Data & Validation

A collaboration born from curiosity, teamwork, and love for the stars.

---

📜 License

This project is licensed under the Apache License 2.0 — free to use, modify, and distribute with attribution.

---

🌎 Acknowledgments

• NASA Exoplanet Archive – for the K2 Planets and Candidates Catalog
  Dataset Link
• Scikit-learn, XGBoost, Flask, Seaborn – core technology stack
• NASA Space Apps Hackathon 2025 – for the opportunity to explore the universe with AI

---

⭐ If you like this project...

Give it a star on GitHub 🌟 and help us reach more open-source astronomers!