Time Series Prediction Based on Generalization Bounds for Support Vector Machine

The fundamental problem of selecting the order and identifying the time varying parameters of an autoregressive model (AR) concerns many important fields. The Vapnik-Chervonenkis (VC) generalization bound provides a mathematical framework for the practical models selection from finite and noisy data sets of time series dataset. In this paper, based on the VC generalization bound for Support Vector Machine (SVM), we introduce a new method of identifying the time varying parameters of an AR model, then and two SVM-based time series prediction models are formulated. Both numerical experiments and theoretical analysis show that the proposed models are feasible and effective.