Attention Recognition Based on EEG Using Multi-Model Classification

Attention recognition based on EEG signals provides a reliable approach for detecting attention states. This study proposes a multi-model classification method to identify three different attention states, such as focused, unfocused, and drowsy. EEG signals are preprocessed by several filters, including Independent Component Analysis (ICA). Three classifiers-SVM, Random Forest, and CNN-are trained individually and combined using fusion strategies such as majority voting and stacking. Results show that Random Forest achieves the highest 920/0 accuracy in individual tests, while the RF + CNN hybrid reaches 96% accuracy with a macro F1-score of 0.93, demonstrating the effectiveness of model fusion for EEG-based attention recognition.