Hybrid classification model for emotion detection using electroencephalogram signal with improved feature set

Abstract

ObjectiveEmotions are complex psychological states that involve a combination of subjective experience, physiological response, and behavioral or expressive response. Emotions are dynamic and can vary greatly between individuals, which poses a challenge for creating models that accurately recognize emotions across diverse populations. There is a significant challenge in generalizing findings from Electroencephalogram (EEG) devices and data sources, which can affect the reliability and applicability of emotion recognition systems. By increasing the precision and dependability of emotion identification systems, this work suggests a hybrid model with an enhanced feature set for emotion detection (HMIFED). MethodsThe Wiener filtering technique is used to preprocess the input EEG data in order to separate the original data from the noisy signal. The preprocessed signal is then used to extract the enhanced entropy-based characteristics, spectral flatness, fluctuation index, and spectral spread features. Using the retrieved characteristics for the emotion classification process, a hybrid model incorporating models such as Bidirectional Long Short-Term Memory (Bi-LSTM) and Improved Recurrent Neural Network (IRNN) is introduced and trained. The classification results are happy, sad, calm, and angry. ResultsComparing the suggested HMIFED model to the traditional methods, the proposed model gets a maximum specificity value of 0.9747. SignificanceThus the proposed model contributes to the field of emotion recognition, which is a critical area of research in human–machine interfaces. The ability to accurately recognize human emotions using EEG signals can greatly enhance the interaction between humans and machines.