Emotion detection using EEG: hybrid classification approach

Abstract

The field of emotion research facilitates the development of several applications, all of which aim to precisely and swiftly identify emotions. Speech and facial expressions are the main focus of typical emotion analysis, although they are not accurate indicators of true feelings. Signal analysis, namely the electroencephalograph (EEG) of the brain signals, is the other area in which emotions are analyzed. When compared to other modalities, EEG offers precise and comprehensive data that facilitates the estimation of emotional states. In order to categories the emotions using an EEG signal, this work suggests a hybrid classifier (HC). The input EEG data is preprocessed using the wiener filtering approach to extract the original information from the noisy signal. The preprocessed signal is used to extract features, such as entropy and a new hybrid model that includes models such as Bi-directional long short-term memory (Bi-LSTM) and improved recurrent neural networks (IRNN), which trains using the retrieved features, is included as part of the classification process. Happy, sad, calm, and angry are the categorization findings; the suggested work demonstrates more accurate classification results than the traditional approaches. All these are done on DEAP dataset with 60%, 70%, 80%, and 90% training sets and also a new DOSE dataset is been created similar to DEAP dataset.