Deep Learning for Skin Cancer Detection: Multi-Class Lesion Classification Using CNN Architecture

Skin cancer ranks among ubiquitous malignancies, its prevalence escalating due to ecological shifts and protracted ultraviolet (UV)exposure. This study aims to address the pressing need for improved diagnostic methodologies by enhancing multi-class skin cancer classification through deep learning techniques, utilizing the HAM10000 ("Human Against Machine with 10,000 training images") dataset. Convolutional Neural Network (CNN) architectures, including VGG16 accuracy-90.43%, VGG19-89.39%, MobileNet-96.75% and EfficientNet B3-99.78%, employed for classification.Data augmentation enhances performance, while transfer learning and feature extraction refine evaluation, enabling EfficientNetB3 to achieve 99.78% accuracy in distinguishing harmful from non-harmful lesions with superior precision, recall, and F1- scores.Findings accentuate the indispensability of resilient CNN architectures and pioneering computational methodologies for augmenting diagnostic precision and reliability.