Enhancing Multi-Class Skin Lesion Classification with Modified EfficientNets: Advancing Early Detection of Skin Cancer

Skin cancer is a prevalent form of cancer among humans, and its diagnosis typically involves screening, dermoscopic analysis, histopathological assessment, and biopsy. Physicians often face challenges distinguishing between similar lesions, leading to time-consuming examinations. In this paper, the EfficientNetsB0-B7 architecture is utilized for multi-class skin lesion classification by leveraging transfer learning (TL) and modifying pre-trained weights from the ImageNet dataset. The empirical evaluation of a publicly available International Skin Imaging Challenge (ISIC2019) dataset highlights the effectiveness of the modified EfficientNetB5 architecture, achieving the highest accuracy, precision, recall, and F-Score in multi-class skin lesion classification. Interestingly, this study reveals that increasing model complexity does not necessarily contribute to improved classification performance. These findings contribute to developing more efficient techniques for identifying skin lesions and combating skin cancer.