In EEG-based emotion recognition, subject’s affective responses are captured by several channels (scalp level) by presenting target stimuli. However, the emotional responses are inconsistent throughout the acquired signal, and rather, it arises at certain duration with high prominence on certain channels. However, existing studies ignored this vital issue and considered the entire acquired signal for processing leading to inaccurate results. Therefore, this work proposes emotion recognition by identifying highly affective EEG segments from automatically chosen relevant channels based on instantaneous phase synchronization measurement among all channels involving Hilbert transform and phase locking values followed by a majority voting algorithm. Next, a random matrix theory discriminates the appropriate segments whose power spectral density has been used in the

$k$

-NN-based classification. Experimental validation has been performed using the EEG dataset from SEED (with positive, neutral, and negative emotions). Five EEG subbands (

$\delta $

$\theta $

$\alpha $

$\beta $

, and

$\gamma$

) are exclusively examined for their high correlation with emotions. The analyses include distinguishing proper segment lengths and their locations, minimum dominant channels while achieving the best classification for individual subband. The proposed method achieved the highest classification accuracy of up to 95% for higher frequency subbands (

$\beta $

and

$\gamma$

) with 15 channels. Furthermore, the segment locations for positive and neutral emotions lie from start till 75% of the entire experiment time, whereas, for negative emotion, it is 25%–75%. Moreover, channels from the left frontal, central, and temporal regions are found very active, which is steady. In a comparative study, the proposed idea demonstrates its superiority by displaying the highest efficiency considering less data from minimum channels.

Parthana Sarma

Publications