A novel ensemble model for multiclass skin cancer classification using deep learning

Skin cancer is a major health concern worldwide, with melanoma being the deadliest form due to its high potential to spread if not detected early. Traditional diagnostic methods, such as visual inspection and biopsy, are time-consuming, invasive, and highly dependent on the expertise of dermatologists. With the advancement of computer-aided diagnosis (CAD) and deep learning, automated classification of skin lesions has become more reliable and accurate. This paper presents a novel ensemble deep learning approach that integrates three pre-trained convolutional neural networks—Xception, ResNet50, and DenseNet201—to enhance the classification of multiple types of skin cancer. The dataset, collected from hospitals and the HAM10000 benchmark, was pre-processed using resizing, normalization, augmentation, and histogram equalization techniques. Experimental results demonstrate that the proposed ensemble model achieves 93% test accuracy, along with 92% recall and 92% F1-score, outperforming the individual models. These findings highlight the robustness of ensemble learning in medical image analysis. The proposed framework has the potential to support dermatologists in clinical decision-making, promote early detection, and contribute to non-invasive, cost-effective skin cancer diagnosis.