Fused CNN-LSTM deep learning emotion recognition model using electroencephalography signals

Abstract

Introduction: The traditional machine learning-based emotion recognition models have shown effective performance for classifying Electroencephalography (EEG) based emotions. Methods: The different machine learning algorithms outperform the various EEG based emotion models for valence and arousal. But the downside is to devote numerous efforts to designing features from the given noisy signals which are also a very time-consuming process. The Deep Learning analysis overcomes the hand-engineered feature extraction and selection problems. Results: In this study, the Database of Emotion analysis using Physiological signals (DEAP) has been visualized to classify High-Arousal- Low-Arousal (HALA), High-Valence-Low-Valence (HVLV), familiarity, Dominance and Liking emotions. The fusion of deep learning models, namely CNN and LSTM-RNN seems to perform better for the analysis of emotions using EEG signals. The average accuracies analyzed by the fused deep learning classification model for DEAP are 97.39%, 97.41%, 98.21%, 97.68%, and 97.89% for HALA, HVLV, familiarity, dominance and liking respectively. The model has been evaluated over the SJTU Emotion EEG Dataset (SEED) dataset too for the detection of positive and negative emotions, which results with an average accuracy of 93.74%. Conclusion: The results show that the developed model can classify the inner emotions of different EEG based emotion databases.