Evaluating the Impact of Optimizer Hyperparameters on ResNet in Hanacaraka Character Recognition

Abstract This study evaluates the performance of various optimizers on a ResNet-18 based Convolutional Neural Network (CNN) model for the task of recognizing Hanacaraka Javanese script characters. The image dataset is divided into three sets: training, validation, and test, with image sizes of 64 × 64 pixels and a batch size of 64. The tested optimizers include SGD, Adam, RMSprop, Adagrad, Adadelta, NAdam, and Adamax, all with a learning rate of 0.001 and trained for 10 epochs. The results show that NAdam provides the best performance with accuracy, precision, recall, and F1-Score values reaching 100 %, followed by Adamax with metrics above 97 %. Adam and Adagrad also demonstrate high performance with metric values above 97 %. Meanwhile, SGD shows fairly good performance with an accuracy of 93.72 %, and Adadelta shows adequate performance with an accuracy of 86.58 %. RMSprop yields the lowest performance with an accuracy of 81.74 %. Accuracy and loss graphs indicate that Adam and Adadelta offer the best balance between training and validation performance, while RMSprop and NAdam exhibit significant instability. This study highlights the importance of selecting the appropriate optimizer to achieve optimal performance in Hanacaraka character classification, with NAdam and Adamax being the best choices.