An algorithm for parametric modelling of a series of time intervals

An algorithm for parametric modelling of a specific type of time series, namely a series of time intervals is proposed and discussed in the paper. The necessary preprocessing steps are presented. They include timebase computation, interpolation, re-sampling and, depending on the application, detrending. The proposed approach of parametric modelling is based on autoregressive moving average (ARMA) model. The methods for ARMA model derivation assume that the model orders (the lengths of autoregressive and moving average filters) are known a priori. Since the result is highly sensitive to the order choice, it is necessary to establish rules for proper order selection. The solution to this, often underestimated, problem is also addressed in the article.