Automated Identification of Exoplanets with Machine Learning

Abstract

The detection of exoplanets is a rapidly evolving field, increasingly supported by advances in Machine Learning. In this work, we explore the capabilities of the AstroNet deep learning algorithm when applied to the light curves preprocessed by the TFAW algorithm. The goal is to classify Threshold Crossing Events (TCEs) and identify new potential exoplanet candidates. We first validate the performance of the model on a subset of previously confirmed exoplanets, showing that the algorithm successfully recovers the expected high prediction scores. Subsequently, we analyze a visually selected subset of 478 candidates from the TFAW survey with assigned priority levels, using the model output to propose priority reclassifications based on objective criteria. Finally, we apply the model to a dataset of 65.970 K2 light curves, identifying 3.800 previously unreported candidates. Our results demonstrate that AstroNet, when combined with TFAW, is a powerful tool for automatic exoplanet candidate classification. However, we also emphasize that such models are not definitive, and complementary validation methods remain essential to confirm the planetary nature of any new transiting candidate.