A machine learning project built for the NASA Space Apps Challenge, designed to detect potential exoplanets using a Random Forest classifier trained on NASA’s Kepler Object of Interest (KOI) dataset.
The dataset contains nearly 10,000 celestial objects, each described by 17 physical and orbital parameters. The model uses these features to predict whether a given object is likely an exoplanet or a false positive, inspired by the classic transit detection method used in astronomy.
Try it out here 👉 Exoplanet Detector Website
You can input the 17 parameters directly on the website and instantly get a prediction on whether the object is an exoplanet or not.
⚠️ Note: The live demo may be taken down in the future as server availability may change.
- Algorithm: Random Forest Classifier
- Dataset: NASA Kepler Object of Interest (KOI) dataset
- Parameters: 17 stellar/orbital parameters per object
- Goal: Classify whether a celestial body is a confirmed exoplanet
- Model Accuracy: 93%
The project includes complete preprocessing, training, and evaluation code, along with a ready-to-use model.
git clone https://github.com/arungjose/exoplanet_koi.git
cd exoplanet_koiWindows:
python -m venv venv
venv\Scripts\activatemacOS / Linux:
python3 -m venv venv
source venv/bin/activatepip install -r requirements.txtpython test_model.pyThis script runs the pre-trained model with a predefined set of parameters for an exoplanet (you can modify them in the script to test your own data).
The project is inspired by the transit photometry technique, where periodic dips in a star’s brightness indicate a planet passing in front of it. The AI model learns from these patterns in the Kepler dataset to automate and refine detection accuracy.
Feel free to clone, experiment, and develop this project further.
If you improve the model, visualization, or UI — contributions are always welcome.
Happy Coding ❤️