Mental Tasks Classification Based on Nonlinear
Parameters
Liu Hailong, Wang Jue, Zheng Chongxun
(Key Laboratory of Biomedical Information Engineering of Education Ministry, Xi'an
Jiaotong University, Xi'an 710049, China)
Abstract: Functions of nonlinear parameters, computed from electroencephalography (EEG)
signals, in mental tasks classification were investigated, where the largest Lyapunov
exponent, the mean period of trajectories and the average initial distance between
neighboring trajectories were taken as the nonlinear parameters, and Fisher¡¯s linear
discriminant was adopted as the classifier. There were a total of 60 task pairs from 4
subjects for classification. The average classification accuracy obtained on 2-second EEG
segments reached to 82.3%, 90.7%, and 93.3% for the above three parameters respectively.
With the third parameter, the average accuracy of 90.8% was achieved on 1-second EEG
segments, which approached favorably to the results of Anderson, et al. The methods of
mean period and average initial distance of trajectories with computationally less
demanding can be used for online analysis.
Keywords: electroencephalography; mental tasks classification; Lyapunov exponent; mean
period; initial distance