An Improved Eeg Emotion Recognition Using Valence Lateralization And Svm-Based Rules Extraction

Pane, Evi Septiana (2019) An Improved Eeg Emotion Recognition Using Valence Lateralization And Svm-Based Rules Extraction. Doctoral thesis, Institut Teknologi Sepuluh Nopember.

Text 0711156002001-Doctoral.pdf Restricted to Repository staff only Download (3MB) | Request a copy

Abstract

Recent advances of human-machine-interaction in emotion recognition are using physiological signals (such as SpO2, ECG, pulse-rate, EEG). Among existing physiological signals, EEG appears to be more reliable to reveal the actual underlying emotional state of the brain. Measuring brain signal by involving EEG is a complicated and delicate work because it includes many features, channels, and frequency bands. An efficient and effective approach for EEG emotion recognition is then becoming crucial. The proposed approach should help to enhance the classification performance of EEG-based emotion recognition. The main objective of this research is an investigation to find the best strategies that can improve the accuracy of EEG emotions recognition. It includes features extraction identification, selecting optimum EEG channels, performing a proper dimension reduction method, incorporating the emotional lateralization, and providing transparent rules of emotion classification. The entire proposed method is implemented on public EEG dataset of emotion with three target classes of positive, negative, and neutral emotions. The results show that differential entropy features can be relied upon for the recognition of emotions from EEG compared to other typical feature extraction methods. In addition to the feature extraction method, the location of the optimal channels for emotion processing is found in the frontal and temporal-lateral regions of the brain asymmetrically. This study also demonstrates the ability of Multiclass Fisher Discriminant Analysis (MC-FDA) to reduce the dimension of the dataset while maintaining its excellent performance of classification accuracy. Another result from employing emotion lateralization shows a significant performance improvement of the classification accuracy by valence lateralization. The result shows that positive emotions are dominant in the left hemisphere, while negative emotions are better recognized from the right hemisphere. Through rule-based classification results, Support Vector Machine (SVM) with Random Forest (RF) achieve the best accuracy and precision compared to other methods. SVM+RF is also able to provide transparent rules for predicting all classes of emotions. The sample rule proves that gamma-frequency bands are highly reliable for emotional recognition. Overall, this proposed strategy has provided evidence on increasing emotion recognition of EEG through several recommendations of feature extraction, channel location, dimension reduction methods, emotional lateralization, and classification methods.

Item Type:	Thesis (Doctoral)
Uncontrolled Keywords:	EEG emotion classification, differential entropy, electrodes selection, multiclass dimension reduction, lateralization, rules-based classifier.
Subjects:	T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK5102.9 Signal processing. T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK5105.546 Computer algorithms
Divisions:	Faculty of Electrical Technology > Electrical Engineering > 20001-(S3) PhD Thesis
Depositing User:	Evi Septiana Pane
Date Deposited:	23 Jul 2026 06:52
Last Modified:	23 Jul 2026 06:52
URI:	http://repository.its.ac.id/id/eprint/70684

Actions (login required)

View Item