EEG-Based Silent Speech Classification Using Native Arabic Silent Speech Dataset with Model Comparison and Optimization

Silent speech recognition using EEG data has emerged as a promising solution for enabling communication in the absence of vocal articulation. This paper presents a comprehensive study of EEG-based silent speech decoding using the Native Arabic Silent Speech Dataset as part of the SSDC 2025 Challenge. We investigate the impact of feature engineering, preprocessing, and model optimization in boosting classification performance. Multiple classifiers, including Random Forest, Support Vector Machine, and K-Nearest Neighbors, were evaluated. Our best-performing model achieved an accuracy of $78.5 \%$ and an F1-score of 0.81. We highlight the effectiveness of preprocessing techniques in enhancing signal quality and classification performance. The results suggest that combining feature engineering and model comparison strategies can significantly improve silent speech decoding using EEG data.