Business processes consist of process activities that must be executed to reach a business goal. The processing times of process activities, as well as the waiting times preceding them, are often influenced by inherent uncertainties, resulting in variability in the overall processing duration of the business process. Current data-driven business process simulation approaches utilize historical data of waiting and activity processing times to fit simple single-peaked probability distributions, from which samples are drawn during the simulation. Such probability distributions might be too simplistic and lead to poor simulation results. Probabilistic learning techniques enable the modeling of uncertainties as non-parametric probability distributions, whose shapes dynamically adapt to influencing factors. This work examines the applicability of a recently proposed probabilistic learner, DR-BART, to express uncertainties of activity processing and waiting times. We train multiple DR-BART models using different combinations of input features on different data sets and sample from these models in a business process simulator. We compare the simulation results with those obtained by sampling from parametric probability distributions. Our results show that DR-BART models can be used to improve business process simulation.