COMPARATIVE PERFORMANCE OF EFFICIENTNET-B0 AND RESNET-50 FOR MELANOMA DETECTION IN DERMOSCOPY IMAGES

Abstract

Melanoma is the most aggressive form of skin cancer with high metastatic potential, and early detection is crucial for improving patient survival. Although deep learning models such as ResNet-50 and EfficientNet-B0 have shown promising results in melanoma classification, systematic comparisons using identical experimental protocols remain limited. This study aims to comprehensively compare the performance of EfficientNet-B0 and ResNet-50 in detecting melanoma from dermoscopy images across multiple evaluation dimensions, including accuracy, precision, recall, F1-score, and computational efficiency. A quantitative experimental research design was employed using the publicly available HAM10000 dataset, consisting of 10,015 dermoscopy images categorized into melanoma and non-melanoma classes. Both models were implemented using transfer learning with ImageNet pretrained weights, trained under identical conditions including data augmentation, class imbalance handling using weighted loss, and standardized hyperparameters. Results showed that EfficientNet-B0 achieved superior performance with 91.5% accuracy, 89.8% precision, 88.2% recall, and 89.0% F1-score, compared to ResNet-50 which achieved 89.2% accuracy, 87.5% precision, 85.3% recall, and 86.4% F1-score. Furthermore, EfficientNet-B0 demonstrated significant computational advantages with only 5.3 million parameters (79% fewer than ResNet-50’s 25.6 million). In conclusion, EfficientNet-B0 outperforms ResNet-50 in both accuracy and computational efficiency, making it more suitable for deployment in resource-constrained clinical environments such as mobile telemedicine applications.