Analysis of frequency domain features for the classification of evoked emotions using EEG signals

• Published in: Experimental brain research Vol. 243; no. 3; p. 65
• Main Authors: Adhikari, Samannaya, Choudhury, Nitin, Bhattacharya, Swastika, Deb, Nabamita, Das, Daisy, Ghosh, Rajdeep, Phadikar, Souvik, Ghaderpour, Ebrahim
• Format: Journal Article
• Language: English
• Published: Berlin/Heidelberg Springer Berlin Heidelberg 01.03.2025; Springer Nature B.V
• Subjects: Adult; algorithms; Biomedical and Life Sciences; Biomedicine; brain; Central nervous system; Classification; domain; EEG; Electroencephalography; Electroencephalography - methods; Emotions; Emotions - physiology; Evoked Potentials - physiology; fearfulness; Female; human physiology; Humans; Male; Neurology; Neurosciences; Photic Stimulation; Research Article; Signal Processing, Computer-Assisted; Visual stimuli; Young Adult; Intrinsic mode function (IMF); Variational mode decomposition (VMD); Electroencephalogram (EEG); Emotion classification; Frequency domain analysis; Machine learning
• ISSN: 0014-4819; 1432-1106; 1432-1106
• DOI: 10.1007/s00221-025-07002-1

Emotion is a natural instinctive state of mind that greatly influences human physiological activities and daily life decisions. Electroencephalogram (EEG) signals created from the central nervous system are very useful for emotion recognition and classification. In this study, EEG signals of individuals are analyzed by the variational mode decomposition (VMD) for frequency domain features to recognize visual stimuli-based evoked emotions (happy, sad, fear). After cleaning EEG signals from artifacts, VMD is employed to decompose the signal into its respective intrinsic mode functions (IMFs). A sliding windowing approach is adopted to calculate the power distributions in each of the predefined frequency bands. The results reveal that extracting frequency domain features using a sliding window of 3 s significantly enhances the efficiency of analyzing induced emotions in subjects. The random forest model shows promising results in classifying various emotions, achieving an accuracy of 99.57% for validation and 99.36% for testing. Moreover, it is observed that the fifth IMF has a strong relationship with emotion elicited from visual stimuli. In addition, the features of the trained model are analyzed by Shapley additive explanations.