Nonlinear time-series modelling by means of self-exciting threshold AR models

Self-exciting threshold autoregressive (SETAR) modelling has been used as a n additional tool in nonlinear analysis of EEG data recorded during typical absence seizures in children suffering from childhood absence epilepsy Fir stly a SETAR model was fitted to the data by means of a novel adaptive esti mation strategy. Secondly the real EEG data were compared with wave-forms g enerated from the resulting model. It was demonstrated that impulse respons es (with infinite length) of the SETAR model reconstructed the TAS pattern with remarkable accuracy. Real and model EEG as well as their SW-shuffled s urrogate signals were compared with respect to the autocorrelation function , correlation dimension, pointwise dimension and the largest Lyapunov expon ent.