Prediction of Neutrino Emissivity from Pre-supernova Stars by Machine Learning

Understanding the pre-supernova phase of massive stars is crucial due to substantial neutrino emissions during various nuclear processes. We developed a machine learning model trained on MESA profiles and corresponding neutrino emission data to predict neutrino spectra for new stellar profiles with high accuracy. Our approach involved preprocessing MESA profile data from 15 M - 30 M stars and training a neural network to map inputs to neutrino emission outputs. The model accounts for four key processes in neutrino formation: pair process, beta process, photo process, and plasma process. The model's predictions were validated against traditional computational results, showing reduced computation time from hours to seconds. This method accelerates prediction and allows real-time monitoring of potential supernova candidates, enhancing our ability to anticipate stellar collapses within 1-3 days before core collapse. Building on works by Patton et al. (2017), our model offers a scalable, efficient predictive tool for neutrino spectra. This advancement promises more rapid, accurate stellar modeling, potentially revolutionizing astrophysical phenomena analysis.