Using a detailed set of hydrodynamic and suspended-sediment observations, we developed data-driven algorithms based on Boosted Regression Trees and Artificial Neural Networks to predict suspended-sediment reference (near-bed) concentration using water depth, wave-orbital semi-excursion, wave period and bed-sediment grainsize as inputs. With one exception, the response of the data-driven algorithms was physically sound; the exception was the response to water depth. Outside of the range covered by the data, predictor performance could not be assessed and is not necessarily reliable. Boosted Regression Trees provide the best predictor of suspended-sediment reference concentration and have a clear explanatory power. Artificial Neural Networks provide slightly poorer predictions. Although the response of the latter is more difficult to interpret, they can be more easily included in numerical models simulating larger (in space) and longer (in time) morphodynamic behavior. Within the range of variability provided by the measurements, these algorithms outperform conventional process-based predictors.