The success of Support Vector Machine (SVM) depends critically on the kernel and the
parameters in it. One of the most reasonable approaches is to select the kernel and the
parameters by minimizing the bound of Leave-one-out (Loo) error. However, the
computation of the Loo error is extremely time consuming. Therefore, an efficient strategy is
to minimize an upper bound of the Loo error, instead of the error itself. In fact, for Support
Vector Classification (SVC), some famous bounds have been proposed. This paper is …

Minimizing bounds of leave-one-out errors is an important and efficient approach for support
vector machine (SVM) model selection. Past research focuses on their use for classification
but not regression. In this letter, we derive various leave-one-out bounds for support vector
regression (SVR) and discuss the difference from those for classification. Experiments
demonstrate that the proposed bounds are competitive with Bayesian SVR for parameter
selection. We also discuss the differentiability of leave-one-out bounds.