🌟 EXOVISION - EXOPLANET DETECTION SYSTEM

NASA Space Apps Challenge 2025 Project

Team: Code Surgeons Innovator Challenge: Exoplanet Detection using AI/ML
Dataset: NASA Exoplanet Archive PHOTOMETRIC Data

🚀 Project Overview

EXOVISION is an advanced machine learning system that automatically detects exoplanets by analyzing stellar light curves from NASA's photometric dataset. Our system uses the transit method to identify the characteristic dimming patterns that occur when an exoplanet passes in front of its host star.

🎯 Key Features

• Multi-Model AI Pipeline: 5 optimized machine learning models (SVM, Random Forest, LightGBM, Gradient Boosting, Ensemble)
• Fast Training: Complete model training in under 22 seconds (15x speed improvement)
• Web Interface: Interactive Flask web application for real-time predictions
• Automated Data Processing: Intelligent feature extraction from light curve data
• NASA Dataset Integration: Direct processing of NASA Exoplanet Archive data
• Visualization Tools: Comprehensive plots for data exploration and model evaluation

🏆 Performance Highlights

Model	Accuracy	F1-Score	Training Time
SVM (Best) ⭐	53.0%	0.329	~4s
LightGBM	61.5%	0.238	~6s
Gradient Boosting	64.5%	0.145	~5s
Random Forest	67.0%	0.057	~4s
Ensemble	63.5%	0.099	~3s

SVM selected as best model due to optimal F1-score for imbalanced exoplanet detection

🛠️ Technology Stack

• Backend: Flask, Python 3.12
• Machine Learning: scikit-learn, LightGBM, TensorFlow
• Data Processing: pandas, NumPy, SciPy
• Visualization: matplotlib, seaborn
• Frontend: HTML/CSS/JavaScript
• Deployment: Local Flask server

📁 Project Structure

EXOVISION/
├── app.py                          # Flask web application
├── exoplanet_detection.ipynb       # Main ML pipeline notebook
├── model_comparison.md             # Model performance analysis
├── README.md                       # This file
├── requirements.txt                # Python dependencies
├── data/
│   └── photometric_dataset.pkl     # Processed NASA dataset
├── models/                         # Trained ML models
│   ├── best_model_fast.pkl
│   ├── fast_svm_fast.pkl
│   ├── fast_lightgbm_fast.pkl
│   └── ...
├── static/                         # Web assets
│   ├── css/style.css
│   ├── js/app.js
│   └── plots/
├── templates/
│   └── index.html                  # Main web interface
└── uploads/                        # User data uploads

🚀 Quick Start

1. Clone & Setup

git clone <repository-url>
cd EXOVISION

2. Create Virtual Environment

python -m venv env
env\Scripts\activate  # Windows
# source env/bin/activate  # Linux/Mac

3. Install Dependencies

pip install -r requirements.txt

4. Run the Application

python app.py

5. Open Browser

Navigate to http://localhost:5000 to access the web interface.

📊 Usage

Web Application

1. Upload Data: Upload CSV files with light curve data
2. Select Model: Choose from 5 trained ML models
3. Get Predictions: Receive instant exoplanet detection results
4. View Analysis: Explore detailed visualizations and confidence scores

Jupyter Notebook

1. Open exoplanet_detection.ipynb
2. Run all cells to reproduce the complete ML pipeline
3. Experiment with different models and parameters

🔬 Scientific Approach

Transit Method Detection

Our system identifies exoplanets using the transit method:

• Analyzes periodic dimming in stellar brightness
• Extracts 15 optimized features from light curves
• Applies robust preprocessing and scaling
• Uses ensemble voting for final predictions

Feature Engineering

Key extracted features include:

• Statistical moments (mean, variance, skewness, kurtosis)
• Periodicity detection via autocorrelation
• Transit depth and duration measurements
• Frequency domain analysis
• Outlier and anomaly detection

🎯 NASA Space Apps Challenge Goals

✅ Automated Detection: AI-powered exoplanet identification
✅ Real-time Processing: Fast predictions on new data
✅ User-Friendly Interface: Accessible web application
✅ Scientific Accuracy: Optimized for astronomical data
✅ Scalable Architecture: Handles large datasets efficiently

📈 Model Optimization

Our fast training approach includes:

• Hyperparameter Optimization: Reduced search space for speed
• Feature Selection: Top 15 most informative features
• Cross-validation: 2-fold validation for efficiency
• Class Balancing: Weighted models for imbalanced data
• Pipeline Integration: Automated preprocessing workflows

🔧 API Endpoints

• GET /: Main web interface
• POST /predict: Upload and analyze light curve data
• GET /models: List available ML models
• POST /batch_predict: Batch processing for multiple files

🌟 Future Enhancements

• Real-time NASA data integration
• Deep learning models (CNN/LSTM)
• Multi-planet system detection
• Advanced visualization dashboards
• Cloud deployment and scaling

👥 Team

Code Surgeons Innovator - NASA Space Apps Challenge 2025 participants dedicated to advancing exoplanet discovery through innovative AI/ML solutions.

📄 License

This project is developed for the NASA Space Apps Challenge 2025. Please refer to NASA's data usage policies for the underlying datasets.

🙏 Acknowledgments

• NASA Exoplanet Archive for providing the photometric dataset
• NASA Space Apps Challenge organizers
• Open-source ML community (scikit-learn, TensorFlow, Flask)

---

👨‍💻 Developer

Noaman Ayub
🌐 Portfolio: noamanayub.netlify.app
💼 LinkedIn: linkedin.com/in/noamanayub
💻 GitHub: github.com/noamanayub

---

🌍 Discovering New Worlds Through Machine Learning 🚀