EEG-based Arabic Silent Speech Classification Using Native Arabic Silent Speech Dataset: A Hybrid Model Combining Random Forest and SVM

This research addresses EEG signal-based Arabic inner speech classification using the Native Arabic Silent Speech dataset. With the challenges of Arabic language complexity and class imbalance, a hybrid approach with Random Forest (RF) and Support Vector Machine (SVM) algorithms and Principal Component Analysis (PCA) feature extraction is proposed. Results show the RF model excels with $68.29 \%$ accuracy, high precision and recall, and is stable across all classes. Meanwhile, SVM only achieved $47.08 \%$ accuracy, and the Ensemble model obtained $57.87 \%$. The RF & SVM hybrid model showed better performance than CNN which experienced overfitting. RF is recommended as the main method for inner speech EEG classification. Future research can develop deep learning techniques and real-time integration to improve the effectiveness of the system in real applications.