Implementation of an Automatic EEG Feature Extraction with Gated Recurrent Neural Network for Emotion Recognition

Abstract

AbstractEmotion is a complicated state that influences one's thoughts and behaviour. Recognizing the emotions of a human being is a major research interest in the affective computing after this pandemic situation and which can be applied in medical related fields to cure physical and mental illness. Early detection of stress helps humans to avoid or prevent many diseases related to it. The development of an emotion recognition system using machine learning algorithms has taken a lot of time and effort for researchers and is less focused with Electroencephalography (EEG) signals because EEG signals are noisy, non-linear, and non-stationary. Deep learning algorithms are the most popular solution due to its ability to see images as data. In this paper, we propose a deep learning framework, Gated Recurrent Unit Emotion Recognizer (GRUER) that can detect human emotion with help of EEG signals. This is achieved by implementing four feature extraction algorithms such as Short-Time Fourier Transform (STFT), Wavelet Entropy, Hjorth and Statistical features on dataset and the feature selection method Principal Component Analysis (PCA) is applied to the extracted features to select most significant features to obtain high-accuracy emotion recognizing model. Keras libraries are used to train the model in an appropriate way so that it is neither overfit nor underfit with the data using the Early-stopping function. The performance of the GRUER model is measured using performance metrics such as accuracy, precision, recall and F1-Score are illustrated in the results. The accuracy of the GRUER is 98% and it is a 3-dimensional model which has valence, arousal and dominance for emotion detection. The model loss obtained by GRUER is 1.12 which is low when compared to other models. Finally, the suggested method and its results show that this proposed method outperforms numerous existing emotion recognition systems.KeywordsEmotion recognitionDeep learningEEG signalsClassificationCNNStressDatasetFeature extraction