<![CDATA[This study aims to improve the performance of Batik Keraton motif classification on an imbalanced dataset through the integration of adaptive class weighting and data augmentation within a transfer learning framework. The dataset consists of 1,799 images across four classes (Kawung, Mega Mendung, Parang, Truntum), preprocessed to 224×224 pixels and split stratifiedly into training, validation, and test sets (80/10/10). Three transfer learning architectures—ResNet50V2, VGG16, and EfficientNetB0—were evaluated with adaptive class weighting and geometric augmentation to enhance minority-class representation. The results indicate that ResNet50V2 with pretrained weights achieved the best performance, reaching a test accuracy of 92.78%, macro precision of 93.13%, macro recall of 92.79%, and a macro F1-score of 92.83%. Adaptive class weighting improved sensitivity toward minority classes, while augmentation contributed to model stability and generalization. These findings demonstrate that combining adaptive weighting and augmentation effectively enhances Batik Keraton motif classification under imbalanced data conditions. ]]>
Copyrights © 2026
