Predicting halo occupation and galaxy assembly bias with machine learning

Understanding the impact of halo properties beyond mass on clustering galaxies (namely galaxy assembly bias) remains a challenge for contemporary models clustering. We explore use machine learning to predict occupations and recover bias in semi-analytic formation model. For stellar-mass selected samples, we train Random Forest algorithm number central satellite each dark matter halo. With predicted occupations, create mock catalogues measure bias. Using range environment properties, find that predictions occupancy variations with secondary are all excellent agreement those our target Internal most important prediction, while plays critical role satellites. Our provided usable format. demonstrate is powerful tool modelling galaxy-halo connection, can be used realistic which accurately expected variations, bias, imperative cosmological analyses upcoming surveys.