MULTI-CLASS EMOTION RECOGNITION FROM EEG USING 3D-CNN AND RNN

Abstract

Emotion recognition from electroencephalography (EEG) signals has attracted considerable attention in recent years, as it offers a non-invasive and objective means of assessing emotional processing. In this paper, we propose a novel hybrid model that combines the power of three-dimensional Convolutional Neural Networks (3D-CNN) and Recurrent Neural Networks (RNN) for emotion recognition from EEG signals. Our model extracts spatiotemporal features using a 3D-CNN and captures temporal dependencies using an RNN, achieving state-of-the-art performance in recognizing emotions such as happiness, sadness, anger, and fear. Emotion recognition is a crit ical aspect of human communication and interaction, and its accurate identification has significant implications in various fields, including psychology, neuroscience, and affective computing. We present a novel method for emotional classification, which is evaluated on the DEAP dataset. Our method achieves high accuracy in binary classification of valence and arousal, with scores of 95.45%, 96.63%, and 97.43%, respectively. In the four-class classification task, our models perform with similar accuracy. However, we note that only four emotion spaces are found with binary classification, whereas 8-class classifi cation is more precise. Therefore, we extend our method to 8-class classification and achieve a promising accuracy of 94.83%.To extract features from the EEG, physio logical, and video signals in the DEAP dataset, we use Fast Fourier Transformation (FFT). We employ the same 3D-CNN+RNN architecture for all four classification models, which contributes to the consistency of our results.Overall, our experiments demonstrate the effectiveness of our proposed method for emotional classification, and provide evidence that it can perform well in both binary and multi-class classification scenarios.