Nonlinear prediction on maximum timings of complex time series

Abstract

We have already proposed a nonlinear modeling method which uses event sizes and event timings. In this paper, we consider availability of our method for continuous time series by predicting occurrence timing of maxima and their sizes from continuous time series. In order to evaluate availability of our scheme, we introduce the prediction accuracy by the following two methods. The first one is to predict continuous time series, using all information of the continuous time series. The second is to extract maxima from continuous time series and apply our proposed modeling scheme to the maxima of time series. Comparing these results, we show that our method has higher predictability if there exists an underlying dynamics of observed complex behavior.