EEG-Based Multi-Class Emotion Recognition using Hybrid LSTM Approach

Abstract

Emotion recognition is a crucial task in human-computer interaction, psychology, and neuroscience. Electroencephalogram (EEG)-based multi-class emotion recognition is a novel approach that aims to identify and classify human emotions by analysing EEG signals. Traditional methods of emotion recognition often face challenges in accurately identifying and classifying human emotions due to their complexity and subjectivity. EEG-based emotion recognition provides a direct and objective measure of three emotional states (positive, neutral, and negative), making it a promising tool for emotion recognition. The proposed hybrid LSTM approach combines the strengths of different traditional machine learning algorithms: Gaussian Naive Bayes (GNB), Support Vector Machine (SVM), Logistic Regression (LR), and Decision Tree (DT). The approach was tested on the EEG brainwave dataset, and LSTM achieved an accuracy of 95%, while the proposed hybrid LSTM-GNB, LSTM-SVM, LSTM-LR, and LSTM-DT models achieved 65%, 96%, 97%, and 96% accuracy, respectively. The contribution of this study is the development of a hybrid LSTM approach that combines the strengths of two different algorithms, resulting in higher accuracy for multi-class emotion recognition using EEG signals. The results demonstrate the potential of the hybrid LSTM approach for real-world applications such as emotion-based human-computer interaction and mental health diagnosis.