Hyperparameter Optimization of MobileNet Architecture on Balinese Mask Classification
I Gede Andika, Ayu Manik Dirgayusari, I Nyoman Arnawan, Made Dona Wahyu Aristana, Dewa Putu Yudhi Ardiana · 2024
This study utilized MobileN et to surpass the accuracy of previous research in the field of Balinese mask image classification. Balinese masks as one of the objects of Balinese cultural heritage must be preserved. Based on the data collected, most of young generation in Bali who has no interest in Balinese masks. The approach in the field of technology is known to have a very effective impact on the younger generation because of its rapid and efficient deployment. To improve accuracy, we analyze hyperparameter combinations such as learning rate, batch size, and dropout. Based on 16 scenarios, we got an accuracy of 98.57% for the combination of learning rate 0.00001, batch size 8, and dropout 40%. This value exceeds previous studies using the VGG16 architecture with an accuracy of 85%.