Multiwavelength cluster mass estimates and machine learning

One emerging application of machine learning methods is the inference of galaxy cluster masses. Often cluster mass predictions are made from observables by fitting or deriving scaling relations; if multiwavelength measurements are available, these scaling relation based estimates are combined into a likelihood. Here, machine learning is used in a simulation to instead directly combine five multiwavelength measurements to obtain cluster masses. Comparisons of the contributions of each observable to the accuracy of the resulting mass measurement are made using model-agnostic Importance Permutation values, as well as by brute force comparison of different combinations of observables. As machine learning relies upon the accuracy of the training set in capturing the observables, their correlations, and the observational selection function, and the training set originates from simulations, a few ways of testing whether a simulation and observations are consistent are explored as well.

PDF Abstract