Electronic Theses and Dissertation

ABSTRAK
Penelitian ini bertujuan mengembangkan model klasifikasi batik Kerawang Gayo berbasis deep learning. Batik Kerawang Gayo merupakan warisan budaya Indonesia dengan karakteristik motif khas dan unik. Namun, pemahaman masyarakat tentang berbagai jenisnya masih terbatas dan belum ada model otomatis untuk mengidentifikasinya. Metode deep learning Convolutional Neural Network (CNN) digunakan untuk mengatasi tantangan dalam mengklasifikasikan beragam motif batik Kerawang Gayo. Dataset yang terdiri dari 5 jenis yang dikumpulkan setelah augmentasi sebanyak 2.250, lalu model CNN dilatih untuk mengenali dan mengklasifikasikan pola-pola tersebut, hasil menunjukkan bahwa batch size berpengaruh signifikan terhadap performa model ResNet-50. Performa terbaik diperoleh dengan batch size 32, mencapai akurasi validasi tertinggi 97%, gap loss dan akurasi training-validasi terkecil, serta fluktuasi stabil. Overfitting menjadi masalah pada batch size 16 dengan gap besar antara loss dan akurasi training-validasi, sedangkan batch size 8 mengalami fluktuasi besar pada loss dan akurasi validasi. Meskipun performa terbaik pada batch size 32, masih ada ruang perbaikan dengan optimasi hyperparameter, regularisation, dan arsitektur model untuk meningkatkan akurasi, presisi, recall, dan F1-score. Pemilihan batch size tepat penting untuk mengoptimalkan ResNet-50, karena batch size terlalu kecil menyebabkan ketidakstabilan, sedangkan terlalu besar berpotensi overfitting. Penelitian ini berpotensi meningkatkan pemahaman masyarakat tentang batik Kerawang Gayo dan promosi budaya Indonesia melalui teknologi deep learning.


Kata Kunci : Dataset, CNN, Deep learning, Kerawang Gayo

ABSTRACT This research aims to develop a classification model for Kerawang Gayo batik based on deep learning. Kerawang Gayo batik is an important part of Indonesian cultural heritage with unique and distinct motif characteristics. However, public understanding of its various types is still limited, and there is currently no automated model to identify them. The Convolutional Neural Network (CNN) deep learning method is used to overcome the challenges in classifying the diverse motifs of Kerawang Gayo batik. A dataset consisting of various types was collected, and then the CNN model was trained to recognize and classify these patterns. The results show that batch size has a significant influence on the performance of the ResNet-50 model. The best performance was obtained with a batch size of 32, achieving the highest validation accuracy of 87%, the smallest gap between training and validation loss and accuracy, and stable fluctuations. Overfitting became an issue with a batch size of 16, with a large gap between training and validation loss and accuracy, while a batch size of 8 experienced large fluctuations in validation loss and accuracy. Although the best performance was achieved with a batch size of 32, there is still room for improvement through optimizing hyperparameters, regularization, and model architecture to enhance accuracy, precision, recall, and F1-score. Selecting the appropriate batch size is crucial for optimizing ResNet-50, as a batch size that is too small can lead to instability, while a batch size that is too large can potentially cause overfitting. This research has the potential to enhance public understanding of Kerawang Gayo batik varieties and promote Indonesian culture through deep learning technology. Keywords: Dataset, CNN, Deep learning, Kerawang Gayo