Back to Volume
Paper: Can we Interpret Machine Learning? An Analysis of Exoplanet Detection Problem
Volume: 527, Astronomical Data Analysis Software and Systems XXIX
Page: 183
Authors: Molina, G.; Mena, F.; Bugueño, M.; Solar, M.
Abstract: The exoplanet detection problem - planets that orbit a star outside our Solar System - has focused on the use of time-consuming manual processes. Today the promising techniques are machine learning methods. However, the lack of interpretability in order to understand what the model does, has hampered the improvement and development of the models. In this work, we study the use of classical machine learning methods for detecting confirmed objects from the Kepler mission. Using metadata from the objects and hand-crafted features from the light curves, our study shows that approximately 93% of the data is correctly detected. The extreme behavior of non-exoplanet objects facilitates the recovery of mostly all of these objects (high recall), however our work presents difficulties with confirmed objects overlapping with the non-exoplanet objects (low precision). Because of this, we provide some insights about where the error could be in order to interpret the learning process of our proposal.
Back to Volume