QPSO-Based Hyper-Parameters Selection for LS-SVM Regression

The selection for hyper-parameters including kernel parameters and the regularization is important to the performance of least squares support vector machines (LS-SVM). The existed parameters selection algorithms, such as the analytical, algebraic techniques and particle swarm optimization (PSO) algorithm, have their own shortcomings. In this paper, the problem of model selection for LS-SVM is discussed. A new method selecting the LS-SVM hyper-parameters is proposed based on the principles of the quantum-behaved particle swarm optimization (QPSO). The feasibility of this method is evaluated on data sets produced by sinc function. Experimental results show that LS-SVM of QPSO-based hyper-parameters selection obtains better generalization capability and has more fast convergence speed than PSO-based hyper-parameters selection.