Abstract
The purpose of this study is to make a quantitative search for possibility of deterministic dynamics in MEG (Magnetoencephalogram) data by analysing the correlation dimensions. The time series data of MEG are analyzed not only by the conventional G-P (Grassberger-Procaccia) algorithm but also by an improved estimator of the correlation dimension proposed by K. Judd to overcome some weaknesses of the G-P method (Physica D, 56, 1992). The improved method utilizes the distribution of the interpoint distances directly, where the dimension can be estimated with multinomial maximum-likelihood estimation. The results show that the correlation dimension of MEG data mostly takes values between 6 and 14. We also discuss influence of low-pass filter and noise upon the estimated dimension.
