Nonlinear Time Series Prediction Using LS-SVM withChaotic Mutation Evolutionary Programming for Parameter Optimization

Abstract

Nonlinear time series prediction is studied by usingan improved least squares support vector machine (LS-SVM) regressionbased on chaotic mutation evolutionary programming (CMEP) approachfor parameter optimization. We analyze how the prediction errorvaries with different parameters (σ, γ) inLS-SVM. In order to select appropriate parameters for theprediction model, we employ CMEP algorithm. Finally, Nasdaqstock data are predicted by using this LS-SVM regression based onCMEP, and satisfactory results are obtained.