An Efficient CNN-Based Model for Balinese Dance Classification with Hyperparameter Optimization

The enthusiasm of international tourists for Balinese culture is not equally reflected in the interest of the local younger generation in traditional arts, particularly Balinese dance. This research seeks to address this disparity by preserving Balinese traditional dance through the integration of advanced technology, specifically by developing an optimized Convolutional Neural Networks (CNNs) model for classifying Balinese dances such as Baris, Barong, Kecak, and others. The proposed model was designed with a lightweight CNN configuration through various optimization processes, including hyperparameter tuning for learning rate, batch size, and total epochs. Since no open dataset exists for Balinese dance classification, a custom dataset comprising images from seven dance classes was collected. Data augmentation techniques such as rotation, flipping, and zooming were applied to enhance model performance. Evaluation results show that the proposed model is cost-effective, with 54.55% fewer parameters than traditional models, while achieving 90.12 % accuracy through hyperparameter optimization. Additionally, the model has been deployed as a web-based application using TensorFlow.js, ensuring easy accessibility for end users. This research supports the preservation of Balinese cultural heritage and contributes to the global dissemination and promotion of Balinese traditional dance through digital technology.