Abstract
This paper addresses the problem of generating multi-step-ahead bandwidth prediction. Variation of bandwidth is modeled as a Nu-Support Vector Regression (Nu-SVR) procedure. A parallel procedure is proposed to hybridize constant and binary Particle Swarm Optimization (PSO) together for optimizing feature selection and hyper-parameter selection. Experimental results on benchmark data set show that the Nu-SVR model optimized achieves better accuracy than BP neural network and SVR without optimization. As a combination of feature selection and hyper-parameter selection, parallel hybrid PSO achieves better convergence performance than individual ones, and it can improve the accuracy of prediction model efficiently.