Exoplanet Classifier 🪐

Machine learning system for classifying exoplanet candidates from NASA's Kepler mission data.

Quick Start

1. Install Dependencies

pip install -r requirements.txt
cd frontend
npm install
cd ..

2. Start the Application

Terminal 1 (API Server):

.\start_api.bat

Terminal 2 (React Frontend):

.\start_frontend.bat

Then open http://localhost:5173 (or the port shown in terminal) in your browser.

---

Features

• Smart Feature Selection: Uses only the 19 most important features for accurate exoplanet detection
• Modern Web Interface: React + TypeScript frontend with interactive visualizations
• REST API: FastAPI backend for predictions and metrics
• Real Performance: ~83% accuracy on unseen data using ensemble machine learning

---

Project Structure

exoplanet-classifier/
├── api/                          # FastAPI backend
│   └── main.py                   # API endpoints
├── frontend/                     # React frontend
│   └── src/                      # Source code
├── data/                         # NASA datasets (KOI, K2, TOI)
├── models/                       # Saved model versions & metadata
├── properly_trained_model.joblib # Main trained model
├── requirements.txt              # Python dependencies
├── start_api.bat                 # API server startup
└── start_frontend.bat            # Frontend startup

---

Most Important Features for Detection

The model uses 19 key features across 4 categories:

Signal Quality (5 features)

• koi_dikco_msky - Difference Image KIC-Centroid Offset (most important!)
• koi_dicco_msky - Difference Image Centroid Offset
• koi_max_mult_ev - Maximum Multiple Event Statistic
• koi_model_snr - Transit Signal-to-Noise Ratio
• koi_dikco_mra - KIC-Centroid Offset (RA)

Flux Centroid (5 features)

• Flux-weighted centroid offsets and errors

Orbital Parameters (5 features)

• Period, depth, duration, radius, impact parameter

Stellar Parameters (4 features)

• Temperature, radius, surface gravity, magnitude

---

Features

React Frontend Pages

• Home - Dashboard with model info and feature overview
• Predict - Individual exoplanet classification with random examples
• Batch Upload - CSV file processing for multiple candidates
• Model Retraining - Train new models and manage existing ones
• Metrics - Model performance analysis and feature importance
• Datasets - Browse KOI, K2, and TOI datasets

API Endpoints

• GET / - Health check
• GET /features - List all features with descriptions
• POST /predict - Make predictions
• POST /predict-raw - Predict using raw dataset rows
• GET /metrics - Model performance metrics
• GET /datasets/{name} - Browse datasets with pagination
• GET /random-example/{dataset} - Get random examples for testing
• GET /models - List all trained models

API docs: http://localhost:8000/docs

---

Model Performance (on Held-Out Test Set)

• Accuracy: 83%
• Precision: 82%
• Recall: 83%
• F1 Score: 82%

Per-Class Performance:

• False Positive: 87% precision, 89% recall
• Candidate: 65% precision, 56% recall
• Confirmed Planet: 86% precision, 92% recall

Classes: Confirmed Planet, Candidate, False Positive

Note: These metrics are calculated on a held-out test set (20% of data) that the model has never seen during training. This represents the true performance on unseen data.

---

License

MIT