Testing a cosmological galaxy simulation with unsupervised machine learning

Simulations enables astrophysicists, usually limited to studying observations, to run numerical experiments and test their theories. Simulation results must, though, be tested against observations in order to check how realistic they are. Previous comparisons of galaxy simulations and observations have considered only one or two features at a time (e.g. GSMFs). Using clustering, an unsupervised machine learning technique, invites a comparison that considers more features at a time, so that all aspects of galaxy formation and evolution are captured concurrently. Our work represents the first time that simulations and observations have been compared in this way. We compare simulated and observed galaxies via the k-means clustering algorithm, evaluating the outcomes we find using stability. Simulated galaxies are taken from EAGLE, and observed galaxies from GAMA. We ensure a consistent selection of 5 features for both simulations and observations. We interpret the clusters we find in the context of theories of galaxy evolution.