SVM-based time series prediction with nonlinear dynamics methods

Abstract. A key problem in time series prediction using autoregressive models is to fix the model order, namely the number of past samples re-quired to model the time series adequately. The estimation of the model order using cross-validation is a long process. In this paper we explore faster alternative to cross-validation, based on nonlinear dynamics meth-ods, namely Grassberger-Procaccia, Kégl and False Nearest Neighbors algorithms. Once the model order is obtained, it is used to carry out the prediction, performed by a SVM. Experiments on three real data time series show that nonlinear dynamics methods have performances very close to the cross-validation ones.