<![CDATA[Abstract. The diversity of consumption fish species in traditional markets often creates difficulties in the identification process, particularly for species with similar morphological characteristics. This study aims to compare the performance of two Convolutional Neural Network (CNN) architectures, namely VGG-16 and MobileNetV2, in classifying 15 species of consumption fish sold at Banyuasri Market, Singaraja. The dataset consisted of 750 images, which were divided into 600 training images and 150 testing images. Furthermore, the training dataset was split into 480 training images and 120 validation images. During the training process, on-the-fly image augmentation was applied using rotation, width and height shifting, zoom, shear, horizontal flipping, and brightness adjustment to improve the model's generalization capability. The training process was conducted in two stages: an initial training phase of 20 epochs followed by a fine-tuning phase of 25 epochs. The experimental results showed that MobileNetV2 outperformed VGG-16, achieving an accuracy of 90.67%, precision of 91.28%, recall of 90.67%, and an F1-score of 90.66%, while VGG-16 achieved an accuracy of 88.00%, precision of 89.03%, recall of 88.00%, and an F1-score of 87.33%. Based on these findings, MobileNetV2 is considered more effective and computationally efficient for consumption fish classification and has strong potential to serve as the foundation for developing web-based and mobile-based fish identification systems.]]>
