<![CDATA[Balinese script is a vital cultural heritage that currently faces preservation challenges due to the limited ability of the younger generation to read and understand it manually. This study aims to develop an automatic transliteration system for Balinese script into Latin characters using a Convolutional Neural Network (CNN) based on transfer learning. The dataset consists of 28 classes with 570 images in each class, resulting in a total of 15.960 images, which undergo preprocessing stages including 224×224 resizing, 1/255 normalization, and augmentation, and are split into an 80:20 training and validation ratio. The proposed method compares three pretrained architectures: VGG16 (variations of 20, 50, and 100 epochs), InceptionV3, and ResNet50 (50 epochs) using the Adam optimizer and a 0.0001 learning rate. Performance is evaluated using accuracy, precision, recall, F1-score, and confusion matrix. The results show that VGG16 achieves the best performance with a consistent accuracy of 0.97 and the highest precision of 0.98 at 50 epochs, while InceptionV3 reaches 0.83 accuracy and ResNet50 scores 0.55. Based on these results, VGG16 is selected as the best model to be implemented in the transliteration system. In conclusion, the pretrained VGG16 CNN architecture is proven to be highly effective and provides superior accuracy for the automatic transliteration of Balinese script images into Latin characters.]]>
Copyrights © 2026
