In order to maximize the performance of accelerators and to make them available for social use, it is essential to improve the control technology of accelerators. Manual tuning has a limit to the number of parameters that can be handled, and it is easy to fall into local maxima. Therefore, we developed a system that automatically tunes multiple parameters simultaneously by using machine learning technology. In this study, we first demonstrated the usefulness of Bayesian optimization by simulating hollow beam transport. After that, we conducted an experiment to tune the ion source by Bayesian optimization. In the experiment, four parameters (gas valve, RF frequency, RF power, and intermediate electrode) were tuned simultaneously, and it was shown that the beam brightness could be tuned as an index in about 1.5 hours.