TRANSFER LEARNING CONVOLUTIONAL NEURAL NETWORKS FOR SKIN CANCER CLASSIFICATION

Abstract

Skin cancer is a widespread disease that requires accurate classification methods. Deep learning models can support the classification of different skin cancer classes. This study evaluates several Convolutional Neural Network (CNN) models for this task. Transfer learning is applied to adapt pretrained CNN architectures to skin lesion classification. The HAM10000 dataset is used for model training and evaluation.Several CNN architectures are compared using multiple performance measures. The evaluation considers macro accuracy, network depth, parameter count, and epoch du- ration.These measures are used to assess classification performance and computational requirements. The results show that most evaluated models achieve satisfactory classification accuracy No clear evi- dence of severe overfitting is observed for most models. Among the evaluated architectures, ResNet18 provides the most suitable results.It achieves a favorable balance between classification performance and computational requirements. Therefore, ResNet18 is identified as the most suitable model for this dataset. The findings support transfer learning for automated skin cancer classification.