Corresponding Author: Gaganjot Kaur
Recognizing human emotions via brain signals is one of the latest advancements in the field of bio-signalling technology. Electroencephalography (EEG) is one of the cost-effective technologies that analyze the biomedical signals generated by a human brain that can be used to interpret human emotions. The primary objective of the paper is to present a multistage emotion classification based on EEG signals. In the initial stage, time and time-frequency-based feature extraction is performed followed by Artificial Bee Colony (ABC) based feature selection in the second stage. Finally, the optimized feature set is used for the training, and classification of the emotions is performed using Long Short-Term Memory (LSTM). Here firefly algorithm is used for the hyperparameters tuning of the classification model. The framework is evaluated against state of art classifiers. Evaluated using the Brainwave dataset, the ABC-F-LSTM model outperforms existing classifiers, achieving a recall of 0.9896. It shows notable improvements of approximately 6.79% over ABC-LSTM, 19.39% over Deep Neural Networks (DNN), and 21.64% over Random Forest, highlighting its superior accuracy and robustness in emotion prediction.