Comparison of ResNet-50 and EfficientNet-B0 Method for Classification of Indonesian Sign Language System (SIBI) Using Multi Background Dataset

This research addresses the comparison between two image recognition methods, namely ResNet-50 and EfficientNet- B0, for the classification of Indonesian Sign Language (SIBI) using a multi-background dataset. The aim is to evaluate the effectiveness of both methods in detecting SIBI sign language and determine the most suitable method in real applications. The results showed that data augmentation significantly improved the performance of both models, especially in terms of accuracy, recall, and F1-Score. The EfficientNet-B0 model with data augmentation at 25 epochs showed the best performance with 98% accuracy, 93% recall, and 85% F1-Score, while ResNet-50 achieved 95% accuracy, 90% recall, and 93% F1-Score. This study emphasizes the importance of data augmentation to improve model generalization ability in complex multi-class classification tasks and identifies EfficientNet-B0 as a more stable and practical choice for field implementation.