Akademik Veri Yönetim Sistemi
Ankara International Congress on Scientific Research-XIII, Ankara, Türkiye, 1 - 03 Mart 2026, ss.812-825, (Tam Metin Bildiri)
ABSTRACT Brain–Computer Interfaces (BCIs) are systems that enable the direct transfer of human brain activities to external devices through the analysis of electroencephalography (EEG) signals, and they have attracted increasing attention particularly in motor imagery–based applications such as rehabilitation, assistive robotic systems, and human–machine interaction. In this study, a hybrid approach that integrates traditional feature extraction methods with modern deep learning techniques is proposed for the classification of EEG signals in motor imagery (MI) based BCI systems. Discriminative spatial features are extracted from EEG signals using the Common Spatial Patterns (CSP) method, and the resulting six-dimensional feature vectors are evaluated using different deep learning architectures. Experimental studies are conducted on the BCI Competition IV-2a dataset, which is widely used in the literature for motor imagery classification. The CSP-based features are provided as inputs to Multilayer Perceptron (MLP), Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM), and Graph Convolutional Network (GCN) models, and their abilities to learn spatial, temporal, and topological information are comparatively analyzed. The results demonstrate that the GCN based approach, which represents EEG channels as nodes and inter-channel relationships as edges, models the brain’s natural topological structure more effectively and achieves superior performance with an average classification accuracy of 79.56% compared to the other architectures. The findings indicate that combining CSP, a well-established and computationally efficient feature extraction method, with graph-based deep learning models provides a more robust and stable classification framework against inter-subject variability. The proposed approach is therefore considered a strong and practical alternative for real-time BCI systems operating on portable hardware platforms.