We theoretically investigate the stochastic decay of persistent currents in a toroidal ultracold atomic superfluid caused by a perturbing barrier. Specifically, we perform detailed threedimensional simulations to model the experiment of Kumar et al. in [Phys. Rev. A 95 021602 (2017)], which observed a strong temperature dependence in the timescale of superflow decay in an ultracold Bose gas. Our ab initio numerical approach exploits a classical-field framework that includes thermal fluctuations due to interactions between the superfluid and a thermal cloud, as well as the intrinsic quantum fluctuations of the Bose gas. In the low-temperature regime our simulations provide a quantitative description of the experimental decay timescales. At higher temperatures, our simulations give decay timescales that range over the same orders of magnitude observed in the experiment, however there are some quantitative discrepancies. In particular, we find a much larger perturbing barrier strength is required to simulate a particular decay timescale (between ∼0.15µ and ∼0.5µ), as compared to the experiment. We rule out imprecise estimation of simulation parameters, systematic errors in experimental barrier calibration, and shot-to-shot atom number fluctuations as causes of the discrepancy. However our model does not account for technical noise on the trapping lasers, which may have enhanced the superflow decay in the experiment. For the intermediate temperatures studied in the experiment, we also observe some discrepancy in the sensitivity of the decay timescale to small changes in the barrier height, which may be due to the breakdown of our model's validity in this regime.