Analisis Pengaruh Batch Size dan Learning Rate pada VGG16 Untuk Klasifikasi Citra Aksara Kaganga

Abstract

This study aimed to develop a character recognition model for Kaganga script using deep learning, leveraging the VGG16 architecture pre-trained on large datasets such as ImageNet. The dataset used consisted of labeled Kaganga script images, which were divided into three parts: training data (70%), validation data (15%), and test data (15%). The model training process involved fine-tuning the last few layers of the VGG16 model, while the earlier layers retained the pre-trained weights. To optimize the model's performance, experiments were conducted by testing various combinations of batch sizes (16, 32, 64) and learning rates (0.1, 0.01, 0.001), resulting in nine different parameter combinations. The model was evaluated using accuracy, confusion matrix, precision, recall, and F1-score metrics on the test data. The experimental results showed that the proper hyperparameter settings significantly affected the model's performance. A batch size of 32 with a learning rate of 0.01 provided the best accuracy across training, validation, and test data. While a batch size of 16 yielded decent results with a learning rate of 0.01, the accuracy on the test data was lower, indicating a tendency toward overfitting with smaller batch sizes. In contrast, a batch size of 64 with a learning rate of 0.01 delivered the best test accuracy of 89.1%, although there was a slight drop in validation accuracy. Based on these results, it was recommended to use a batch size of 32 or 64 with a learning rate of 0.01 for the Kaganga script classification task using the VGG16 model.