An enhanced convolutional neural network for classification of Melanoma skin cancer

Abstract

Melanoma is one of the most aggressive forms of skin cancer and accounts for a disproportionately high number of skin cancer–related deaths despite its relatively low incidence. Early and accurate diagnosis is therefore essential for improving patient survival rates. This study presents an enhanced Convolutional Neural Network (CNN)-based approach for automated melanoma classification using dermoscopic images. The proposed model was developed using the HAM10000 dataset, which contains 10,015 labeled dermoscopic images categorized into melanoma and non-melanoma classes. Comprehensive preprocessing techniques, including image resizing, normalization, and data augmentation, were employed to improve model generalization and reduce overfitting. The CNN architecture integrates convolutional and pooling layers for hierarchical feature extraction, dropout layers for regularization, and a Softmax output layer for classification. The model was trained using the Adam optimizer with a learning rate of 0.001, a batch size of 32, and 100 epochs, employing an 80/20 training–testing split with 10-fold cross-validation. Experimental results show that an enhanced CNN achieved an accuracy of 96.5%, with precision, recall, and F1-score values of 96.9%, 96.0%, and 96.4%, respectively. The model outperformed baseline CNN models and selected pre-trained architectures such as ResNet50. These results demonstrate the robustness and effectiveness of the enhanced approach, highlighting its potential for integration into computer-aided diagnostic systems to support dermatologists in early melanoma detection.