Identification and Classification of Exoplanets Using Machine Learning Techniques

NASA's Kepler Space Telescope has been instrumental in the task of finding the presence of exoplanets in our galaxy. This search has been supported by computational data analysis to identify exoplanets from the signals received by the Kepler telescope. In this paper, we consider building upon some existing work on exoplanet identification using residual networks for the data of the Kepler space telescope and its extended mission K2. This paper aims to explore how deep learning algorithms can help in classifying the presence of exoplanets with less amount of data in one case and a more extensive variety of data in another. In addition to the standard CNN-based method, we propose a Siamese architecture that is particularly useful in addressing classification in a low-data scenario. The CNN and ResNet algorithms achieved an average accuracy of 68% for three classes and 86% for two-class classification. However, for both the three and two classes, the Siamese algorithm achieved 99% accuracy.