Light-Weight Gated Recurrent Unit for Electroencephalogram Emotion Classification

This experiment investigates the impact of reducing GRU units and dense layer neurons in a lightweight GRU architecture (LW-GRU-RU) compared to a baseline GRU model for electroencephalogram (EEG) emotion classification. A baseline GRU model is used as a reference, with an optimized GRU variant utilizing feature selection through a Random Forest-based algorithm. Experiments are conducted on a labeled emotion dataset, comparing accuracy and training efficiency across five trials. Results highlight the trade-off between model complexity, accuracy, and computational efficiency, providing insights for practical applications. Both models are evaluated on an emotion dataset for accuracy and training efficiency. The lightweight model achieves a competitive accuracy of 97.486% while reducing the average training time to 0.19 seconds per epoch, showcasing its potential for efficient real-world applications.