Performance Analysis for EEG-Based Emotion Classification Using Deep Learning Algorithms

Electroencephalogram (EEG)-based emotion recognition is a critical area in affective computing and human-computer interaction. This study investigates how well three Long Short-Term Memory (LSTM)-based architectures-Coupled Gate LSTM, Peephole LSTM, and Bidirectional LSTM (BiLSTM)-classify emotion states from EEG data. The dataset consists of Electroencephalogram frequency bands and power spectral density features, preprocessed using standardization and one-hot encoding. The models were assessed according to their performance measures after being trained for 50 epochs using the Adam optimizer with categorical crossentropy loss. Experimental results indicate that BiLSTM provides superior accuracy but at a higher computational cost, while Peephole LSTM improves state tracking, and Coupled Gate LSTM offers a balance between complexity and performance. The study suggests future improvements through attention mechanisms and transformer-based models for enhanced feature selection.