Exoplanet Detection with Decision Trees

Abstract

Exoplanets can be detected through the observations of brightness and movement of the stars they orbit. In the past, machine learning algorithms have been able to classify possible candidates using specific techniques such as analyzing large samples of data and automating the otherwise tedious process of classification. In our research, we train a decision tree algorithm on datasets containing confirmed exoplanets, candidates, and false positives from the Kepler Mission in the NASA Exoplanet Archive. From this training, we build a decision tree classification model with a 94.12% accuracy at classifying exoplanets when training on confirmed exoplanets, candidates, and false positives, and a 99.78% accuracy when training only on confirmed exoplanets and false positives. Alternatively, when training a decision tree regression model to predict Kepler Object of Interest KOI) scores, we obtain a loss of 0.04. The decision tree algorithm is a viable option in classifying and detecting exoplanets, as displayed by its effectiveness.