Data-driven classification of metal-poor stars using machine learning

We present a first-time application of machine learning (ML) to classify metal-poor stars by the astrophysical processes responsible for enriching their stellar environments. The current convention refers to stellar sites as having been enriched by the rapid (r-), intermediate (i-), or slow (s-) neutron capture processes based on simple threshold values of abundance ratios from a very small and potentially restrictive set of elements. In this work, we develop data-driven classifiers trained on nucleosynthesis calculations from simulations of r- and s-process sites. We present the ML classification results and compare them to their conventional categorizations. The elements that play a dominant role in the classification results are highlighted by examining the feature importance. We further discuss additional insights - and some challenges - revealed by this novel approach.