Novel Algorithm for Measuring the Complexity of Electroencephalographic Signals in Emotion Recognition

Emotion recognition in electroencephalographic signals has attracted increasing attention in medical literatures. The Lempel-Ziv complexity (LZC) measurement is an effective method for processing nonlinear dynamical analysis of electroencephalograms (EEGs), and has been used for emotion recognition in affective brain -computer interface (aBCI) systems. In this paper, an improved algorithm based on conventional LZC was proposed for resolving emotion recognition in EEGs. In the new method, the EEG signal was first preprocessed by employing wavelet packet transform to remove EEG series with low frequencies. Then, a nonlinear filter was used to remove ectopic values from the EEG. Furthermore, an effective adaptive LZC algorithm was proposed for measuring the complexity of EEG signals, and this method was applied to recognize emotions. The fluctuations of LZC sequences in designated electrode sites were constructed using three different LZC algorithms. Then, statistical analysis was carried out for group comparisons of the statistically significant fluctuation of LZC sequences. Finally, it was used as the classified features in a support vector machine classification algorithm. The experiment demonstrated that the proposed method effectively discovered more patterns occurring in EEG signals. Also, it precisely detected oscillations of EEG signals, which may help reveal intrinsic nonlinearity in EEG -based emotion recognition. The adaptive LZC algorithm reached a high average accuracy of 82.43%.