Classifying exoplanets as potentially habitable using machine learning

Abstract

With the launch of the ESA Gaia satellite observatory and the planned LSST, and the torrent of data coming through the Kepler space observatory, scientists will be able to collect data for more than 1 billion astronomical objects, including millions of exoplanets in the coming years. In this study, several predictive models are built using machine learning algorithms to classify exoplanets as potentially habitable, based on various characteristics of the planet and its star. I applied six supervised learning algorithms for the classification of planets, which include two decision trees, CART and Random Forest, Support Vector Machines, Logistic Regression, Feed-Forward Neural Network, and Naïve Bayes. I further applied CART to create a regression model to predict the value of the ESI (Earth Similarity Index) for an exoplanet.