Time series modeling of heart rate dynamics

Abstract

Autoregressive (AR), autoregressive-moving average (ARMA), bilinear (BL), and polynomial autoregressive (PAR) models were fit to heart rate time series obtained from 9 subjects in the supine position. For each data set and model structure, model order was determined by the Akaike Information Criteria (AIC). For all data sets, the nonlinear BL model had a lower residual variance and AIC compared to AR models. In most cases, BL models provided a better fit to the data than either ARMA or PAR models. For most data sets, the nonlinear BL model provides a more accurate representation of HR dynamics compared to the other model structures tested. >