ExoDiscover | NASA Space Apps 2025

Exoplanet classification system using machine learning trained on NASA datasets (Kepler, TESS) for the NASA Space Apps Hackathon 2025

Authors

Daniel Kwan, Karl Keshavarzi, Wen Ting Wang, Aadhya Anand

Demo

Option 1: Run the Website Locally

This starts the Flask backend + the Vite frontend.

Backend setup:

cd exodiscover-backend
python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate
pip install -r requirements.txt

Start the API:

python app.py

Start the frontend (in a second terminal):

npm install
npm run dev

Open the app using the given localhost server.

Option 2: Run Predictions via Script

This runs a script-driven predictor and does not use the web app.

Ensure the Python environment is ready (same as Option 1), then run:

python exodiscover-backend/predict.py

ML Component

Data source: NASA Kepler Objects of Interest (KOI) catalog

Features extracted from light curves:

• Statistical: mean, std, median, max, min, range, coefficient of variation
• Transit detection: num_dips, dip_fraction, max_dip_depth, transit_count
• Signal analysis: skewness, kurtosis, SNR, smoothness
• Periodicity: num_periods, primary_period

Target labels:

• Confirmed (verified exoplanet)
• Candidate (potential exoplanet)
• False Positive (not an exoplanet)

Models implemented using scikit-learn and XGBoost:

• XGBoost Classifier (production)
• CNN for light curve analysis
• LSTM for sequential data

Training and evaluation:

python exodiscover-backend/train_final_model.py
python exodiscover-backend/predict.py

Evaluation metrics:

• Accuracy on holdout test set
• Classification report (precision, recall, F1)
• Confusion matrix

Tech Stack

Frontend:

• React
• TypeScript
• Vite
• Tailwind CSS
• shadcn-ui

Backend:

• Flask
• XGBoost
• scikit-learn
• TensorFlow/Keras
• lightkurve (NASA light curve analysis)

Data Sources

• NASA Exoplanet Archive
• Kepler Objects of Interest (KOI) catalog
• Kepler, K2, and TESS mission data