<![CDATA[Facial emotion recognition is one of the key areas of artificial intelligence that has been widely applied in human–computer interaction. This study compares three clustering algorithms—K-Means, Gaussian Mixture Model (GMM), and Spectral Clustering—to group facial expressions from the FER-2013 dataset, which was sampled down to 1,820 images. Facial features were extracted using Gabor filters and reduced in dimensionality using Principal Component Analysis (PCA) to efficiently preserve essential information. The evaluation was conducted using Silhouette Score, Davies–Bouldin Index, and clustering accuracy estimation. The results show that Spectral Clustering achieved the best performance, with a Davies–Bouldin Index of 0.464 and an accuracy of 95.36%, followed by GMM (Silhouette Score 0.526) and K-Means (Silhouette Score 0.523). Furthermore, PCA with 80% variance retention produced an effective 1D feature representation, allowing clustering results to be visualized in a simple yet informative manner. These findings suggest that the choice of clustering algorithm should be aligned with the desired trade-off between system accuracy and efficiency.]]>
Copyrights © 2025
