Nonlinear Granger Causality using ANFIS for identification of causal couplings among EEG/MEG time series

摘要

Identifying the causal couplings among EEG/MEG time series is an important problem in the neuroscience field. For linear stochastic models, Granger causality (GC) is used as a simple concept to explore such interactions. In this paper, we extend GC concept to a nonlinear version based on the Adaptive Neuro Fuzzy Inference System (ANFIS) and propose a new effective connectivity measure (ANFISGC) with capability in detecting linear and nonlinear causal information flow between time series. We applied the proposed method to the simulated datasets and compared its performance with the classic Linear Granger Causality (LGC). In a linear (AR) simulation model, LGC performs the same as ANFISGC but in the case of nonlinear models, ANFISGC outperforms LGC.