A Hybrid CNN-SVM Approach for Skin Lesion Classification Using The ISIC 2019 Dataset

Identifying skin lesions accurately is necessary for immediate diagnosis of skin cancer, particularly melanoma, which can significantly reduce mortality rates. However, automating the categorization of skin lesions poses challenges due to subtle variations in their appearance on the skin. This article presents a hybrid approach that combines Support Vector Machines (SVM) for categorization with Convolution Neural Networks (CNN) for feature extraction. Random Forest (RF) and Decision Tree (DT) methods are also included for comparative analysis. Using ISIC 2019 dataset encompassing dermoscopic images, the evaluation results show that the CNN-SVM hybrid paradigm surpasses other classifiers for accuracy, precision, recall, and F1 score. Experimental findings indicate performance improvements of 3.6% and 6.9% for CNN-SVM model compared to the CNN-RF and CNN-DT models, respectively. These results emphasize the ability of hybrid methods to enhance classification efficiency for medical image analysis tasks.