WATSON-Net: Vetting, Validation, and Analysis of Transits from Space Observations with Neural Networks

The introduction of WATSON-Net marks a significant advancement in the field of exoplanet research, particularly as the number of detected transiting exoplanet candidates continues to rise. This open-source neural network classifier is designed to automate the vetting and validation process, which is critical for prioritizing these candidates. Trained on Kepler Q1-Q17 DR25 data, WATSON-Net achieved an impressive recall-at-precision of 0.99, ranking second only to the ExoMiner network. Additionally, it has been calibrated for TESS data, demonstrating a recall of 0.76 and a precision of 0.93, positioning it as a leading non-fine-tuned machine learning classifier for TESS signals. The development of such robust tools is essential as it enables researchers to efficiently analyze and validate the increasing volume of data from space-based missions, ultimately enhancing our understanding of exoplanets.