Towards model-independent mode detection and characterisation of very long biomedical time series
Pauwels, Karl and Gautama, Temujin and Mandic, Danilo and Van Hulle, Marc M.
Proceedings of the 4th International Conference on Recent Advances in Soft Computing (RASC), The Nottingham Trent University, pp. 230–236, 2002
A novel technique, the Delay Vector Variance method, which provides characterisation of time series in terms of their predictability is introduced and applied in a biomedical context. The method evaluates the predictability, in a robust, model-independent manner, enabling a wide range of applications. The merits of the procedure are demonstrated in a mode segmentation context both analytically and experimentally on a set of long nonstationary physiological signals, obtained from subjects undergoing different sleep and wake stages. It is shown that the features extracted by the method remain consistent, not only over different naps of a single subject, but also allow for across-subject generalisation. Next, the presence of nonlinearity associated with the different modes is investigated. A comparison with other measures supports the obtained results, namely that the signals show a higher degree of nonlinearity during wake than during sleep stages.