Adaptive Feature Optimization with Dynamic GRU for EEG Emotion Classification

This experiment evaluates the effectiveness of a novel adaptive feature optimization (AFO) mechanism integrated with a Dynamic GRU model (DGRU) for emotion classification. The A FO-DGRU model dynamically s elects and adjusts input features during training based on their gradient importance and model performance metrics. Compared to the baseline GRU, this adaptive model achieves higher accuracy at 98.274%, while maintaining computational efficiency with an average training time of 0.25 seconds per epoch. These results highlight the potential of combining adaptive feature selection with reducing GRU architectures for efficient sequence modeling.