High-entropy alloy (HEA) electrocatalysts offer a vast composition space that awaits exploration to identify interesting materials for energy conversion reactions. While attempts have been made to explore the composition space of HEA thin-film libraries and compare experimental and computational studies, no corresponding approaches exist for HEA nanoparticles. So far, catalytic investigations on HEA nanoparticles are limited to small sets of individual catalysts. Here, we report the experimental exploration of the composition space of carbon-supported Pt-Ru-Pd-Rh- Au nanoparticles for the H-2/CO oxidation reaction by constructing a dataset using Bayesian optimization as guidance. Applying a surfactant-free synthesis platform, a dataset of 68 samples was investigated. By constructing machine learning models, the relationship between the concentrations of the constituent elements and the catalytic activity was analyzed and compared to density functional theory calculations. The machine learning models confirm findings from previous studies concerning the role of Ru in the H-2/CO oxidation reaction. This has been achieved starting from a random set of compositions and without any prior assumptions for the reaction mechanism nor any in-depth design of the active site. In addition, by comparing the trends of the computational and experimental studies, it is seen that the "onset potentials " across the compositions can be correlated with the adsorption energy of *OH. The best correlation between the computational and experimental data is obtained when considering 5% of the most strongly *OH adsorbing sites.