Learning the parameters for least squares support vector machine

The regularization parameter and kernel parameter play important roles in the performance of the least squares support vector machine (LS-SVM). Aimed at optimizing the LS-SVM´s parameters, a fast method based on distance is presented. The method is by way of calculating the various types of distances in the feature space to determine the optimal kernel parameter. Since the method only needs to calculate some simple mathematical formulas, and avoids training the corresponding LS-SVM classifiers, the method can greatly reduce the training time. Experiment results show that the proposed method can improve the training speed.