Parameter selection for primal least squares twin support vector regression

Parameter selection is considered crucial for regression models. When we address the parameter selection within primal least squares twin support vector regression (PLSTSVR), an optimization problem defined as the minimization of the testing loss is proposed. In contrast to the traditional grid search or other heuristic methods, our method offers more optimal parameters for PLSTSVR theoretically. Experimental results demonstrate that our method achieves higher prediction accuracy compared to many state-of-the-art parameter selection methods. This parameter selection method can be applied to many other learning machines to improve their performance.