Nonlinear system identification using Relevance Vector Machines (RVM). The ARX-RVM yields confidence intervals for the predictions and provides a better trade-off between accuracy and sparsity.