In recent years, the development of complex multi-metallic nanomaterials like high entropy alloy (HEA) catalysts has gained popularity. Composed of 5 or more metals, the compositions of HEAs exhibit extreme diversity. This is both a promising avenue to identify new catalysts and a severe constraint on their preparation and study. To address the challenges related to the preparation, study and optimization of HEAs, machine learning solutions are attractive. In this paper, the composition of PtRuPdRhAu hydrogen oxidation catalysts is optimized for the CO oxidation reaction. This is achieved by constructing a dataset using Bayesian optimization as guidance. For this quinary nanomaterial, the best performing composition was found within the first 35 experiments. However, the dataset was expanded until a total of 68 samples were investigated. This final dataset was used to construct a random forest regression model and a linear model. These machine learned models were used to assess the relationships between the concentrations of the consituent elements and the CO oxidation reaction onset potential. The onset potentials were found to correlate with the composition dependent adsorption energy of *OH obtained from density functional theory. This study demonstrates, how machine learning can be employed in an experimental setting to investigate the vast compositional space of HEAs.