Evaluation of a nonlinear prediction algorithm quantifying regularity, synchronization and directionality in short cardiovascular variability series

Abstract

An unifying approach evaluating complex dynamics and dynamical interactions in short bivariate time series is presented. The method performs nearest neighbor local linear prediction to estimate regularity, synchronization and directionality of two interacting time series. It was implemented through a specific cross-validation procedure which allowed an unconstrained embedding of the series and a full exploitation of the available data to maximize the accuracy of prediction. The approach was evaluated by simulations of stochastic (autoregressive processes) and deterministic (Henon maps) models in which uncoupled, unidirectionally coupled and bidirectionally coupled dynamics were generated. The method was then applied to representative examples of heart period and systolic pressure variability series, showing its ability to describe complexity and interactions in short term cardiovascular variability.