An information-theoretic measure for the classification of time series

Abstract

We present a practicable procedure which allows us to decide if a given time series is pure noise, chaotic but distorted by noise, purely chaotic, or a Markov process. This classification is important since the task of modelling and predicting a time series with neural networks is highly related to the knowledge of the memory and the prediction horizon of the process. Our method is based on a measure of the sensitive dependence on the initial conditions which generalizes the information-theoretical concept of Kolmogorov-Sinai entropy.