Exoplanets detection using Auto-Tuned Random Forest
Abstract
The NASA Kepler project consists of discovering exoplanets (i.e. planets outside our solar system). The data generated by the Kepler Space Telescope is analyzed by humans and algorithms to discover new exoplanets. The main problem is that a lot of exoplanets are revealed false positives. This work consists of identifying exoplanets using random forest, a Supervised Machine Learning model. Furthermore, the fitted model is analyzed in order to determine which features are relevant. The hyperparameters are automatically optimized with techniques that come from the AutoML research. In fact, the hyperparameters of the model are optimized and cross-validated with Hyperband, a simple yet effective and scalable method for hyperparameters optimization.
Files and Directories description
- requirements.txt contains the Python dependencies used.
- nasa_koi_planets.csv contains the NASA exoplanets dataset (CSV format).
- src/ contains the Python source code used to build, fit and test the model comparing it with other models.
- results/ contains the readable test results of the models (TXT format).
- doc/ contains the LaTeX documentation and a Makefile to build it. It also contains a PDF format of the documentation.