EmHM: A Novel Hybrid Model for the Emotion Recognition based on EEG Signals

Abstract

Emotion is a fundamental aspect of daily life and is crucial for human interactions. This study suggests a unique electroencephalogram (EEG)-based technique for identifying human emotions. For EEG-based emotion analysis, the proposed model is tested on the SEED and DEAP datasets. For the DEAP dataset, we consider valence and arousal emotions for classification purposes, and for the SEED dataset, three emotions, neutral, positive, and negative, have been considered. The differential entropy (DE) is used for the SEED dataset, and for the DEAP dataset, the power spectral density (PSD) is used as a feature. For precise emotion recognition, an EmHM (Emotion Hybrid Model) based on long short-term memory (LSTM) and a convolutional neural network (CNN) are constructed. Furthermore, we applied the CNN, LSTM, and EmHM models, and all three models for emotion recognition are fed with the retrieved information. Various methods improved on already-existing models to accurately classify human emotion. To get better accuracy than the existing techniques, we suggested a model that uses a different approach known as EmHM. By applying all three models such as CNN, LSTM and EmHM, we got the highest accuracy 86.50%, 87.98% and 91.56% respectively on dataset. To improve prediction results, the CNN and the LSTM models are combined to make the EmHM hybrid model.