On the selection of autoregressive order for electroencephalographic (EEG) signals

In the present work, the choice of autoregressive model order is investigated for the analysis of EEG signals. The Akaike Information Criterion gave very disperse results in all of the cases tested-rendering it of little use for the choice of model in such signals. However, certain spectral parameters commonly used in monitoring the EEG were found to be robust to changes over a wide range of orders. It is concluded that while the most suitable model order is difficult to identify, this may not be of great importance to the final result of quantitative EEG analysis. When distinct spectral peaks are of interest, however, higher model orders than commonly recommended are necessary