<![CDATA[The classification of ethnic groups in Indonesia based on facial images faces significant challenges due to high morphological diversity and the limitations of existing computational methods in handling local ethnic variations. This research developed a hybrid classification system to address this problem. The system was built through several stages: collecting a primary dataset of 550 facial images from five ethnic groups (Acehnese, Batak, Florenese, Javanese, and Papuan), extracting facial features using the FaceNet (InceptionResnetV1) model to generate face embeddings, and classification using a Support Vector Machine (SVM). To achieve maximum precision, the SVM model's hyperparameters were automatically tuned using Bayesian Optimization. The model's capabilities were confirmed using an 80/20 training-testing split, resulting an impressive 94.55% of accuracy. Its high discriminative power was further solidified by a stellar 0.9930 AUC-ROC score. Closer inspection showed a fascinating dichotomy: the model pinpointed the Papuan ethnicity with perfect precision, though it occasionally faltered when faced with the subtle morphological overlaps found in other ethnic groups. This study demonstrates that the combination of deep learning feature extraction with an optimized SVM classifier is an effective and robust approach for complex ethnicity classification, successfully providing an accurate and objective classification solution.]]>
Copyrights © 2026
