<![CDATA[Batik plays a crucial role in Indonesian cultural heritage, with regional motifs encoding local philosophies, identities, and socio-historical contexts while also sustaining creative industries and tourism. Automated classification of batik by region can support documentation, education, and authentication, yet remains challenging due to visually overlapping patterns, high intra-class variability, and subtle inter-regional differences. Building on recent advances in Vision Transformers, this study investigates the Swin Small Transformer architecture for classifying Indonesian batik motifs into five regional categories: Jawa Barat, Jawa Tengah, Jawa Timur, Madura, and Yogyakarta. The proposed framework employs the swin_small_patch4_window7_224} model initialized with ImageNet-pretrained weights and fine-tuned on a curated regional batik dataset. The hierarchical shifted-window attention mechanism of Swin is leveraged to capture both local repetitive elements and broader compositional structures that characterize regional styles. Experimental evaluation on a held-out test set consisting of 80 images demonstrates outstanding performance. The model achieves perfect classification results with overall accuracy, macro-averaged precision, recall, and F1-score all reaching 1.0000. No misclassifications are observed across any regional category, indicating that the proposed architecture effectively learns discriminative representations of regional batik motifs. These findings suggest that hierarchical Vision Transformers can robustly model the nuanced visual cues underpinning regional identity in batik patterns and provide a strong alternative to conventional convolutional neural network approaches. Beyond batik classification, the proposed framework may be extended to other cultural-heritage textile applications, supporting digital preservation, educational initiatives, and large-scale documentation of traditional artistic assets.]]>
