Akademik Veri Yönetim Sistemi
2025 Innovations in Intelligent Systems and Applications Conference, ASYU 2025, Bursa, Türkiye, 10 - 12 Eylül 2025, (Tam Metin Bildiri)
Emotion recognition studies based on EEG data hold significant importance in fields such as human-computer interaction and mental health monitoring. However, the complex structure of EEG signals makes it challenging to accurately identify emotional states using traditional methods. In this study, advanced graph neural network models are proposed to effectively and accurately classify emotions by analyzing multi-channel signals obtained from EEG data. The proposed models aim to classify positive, negative, and neutral emotions by evaluating the relationships between EEG channels through a graph-based structure. The graph neural network models used include a convolution-based graph network (GraphGCN), an attention mechanism-based graph network (GraphGAT), and a scalable and inductive learning-based graph neural network (Graph- SAGE). Experimental results show that the GraphSAGE model achieved the best performance with 92% accuracy, while the GraphGCN model achieved 91% accuracy, and the GraphGAT model achieved 89% accuracy in classifying positive, negative, and neutral emotions. These results demonstrate that graph neural network-based approaches provide an effective solution with high accuracy for emotion recognition using EEG data, highlighting their potential in this field.This study is among the first to directly compare multiple GNN architectures for EEGbased emotion classification using graph structures optimized via the k-nearest neighbors algorithm. In this way, the proposed approach contributes to the literature both in terms of accuracy and explainability in EEG signal classification.