Automatic real-time emotion recognition based on GSR and ECG signals becomes an effective computer-aided tool for emotional recognition as a challenge to pattern recognition. Traditional machine learning methods require the development and extraction of various features dependent on extensive domain knowledge. As a result, non-domain experts can find these methods challenging. On the other hand, deep learning methods have been widely used in several current studies to learn features and identify various types of data. In this paper, to characterize human emotion states, we proposed a hybrid neural network that combines ‘Convolutional Neural Network (CNN)’ and ‘Long-Term Short-Term Memory (LSTM)’. Our dataset consists of four types of emotions which are happy, sad, fear, angry. We have trained our model with CNN-LSTM. Our proposed CNN-LSTM model gives 100% training accuracy and 99.05% validation accuracy with RMSProp optimizer. We also compare our result with machine learning algorithms: Random forest, Logistic Regression, Support Vector Machine, and Naïve Bayes. The comparison result clearly shows that our proposed CNN-LSTM gives the best result among the other classifiers.
ISBN för värdpublikation: 978-981-16-7596-6, 978-981-16-7597-3