Electroencephalography-based Human Emotion Detection using Hybrid ABC-F-LSTM Model

The advancement of human-computer interfaces (HCIs) has significantly impacted accurate emotion detection using electroencephalography signals. Existing methods unveil several limitations, including feature redundancy and reduced estimation accuracy for new users when utilizing pre-trained models. Therefore, this study emphasizes improved EEG-based emotion recognition by developing a Swarm Intelligence (SI) based model for feature selection. The proposed model integrates a feature extractor, feature selector, and label classifier interface. It employs time-domain (TD) and time-frequency domain (TFD) analyses to establish fundamental information for emotion recognition. Then, an Artificial Bee Colony (ABC) algorithm will be used to select the most discriminative features from the brainwave data. The Long short-term memory (LSTM) is utilized to train and classify emotions. Another swarm-based algorithm, the firefly algorithm, is used for the tuning of hyperparameters of a classification model. The proposed approach demonstrates a significant improvement in accuracy over ABC-LSTM, LSTM, DNN, RF, and NB at 8.71%, 12.7%,20.2%, 22.7%, and 20.1%, respectively. The proposed hybrid ABC-F-LSTM model with precision 96.39, a recall of 98.86, an F-measure of 97.65 and an accuracy of 92.03% performed best against state-of-the-art classifiers and represents a very effective strategy for EEG-based emotion classification.