Advances in Skin Cancer Detection Using Machine Learning: Current Methods and Future Directions

Abstract

With increasing incidence rates, high mortality risks, and substantial cost burdens, skin cancer is a serious global health concern. In order to improve patient outcomes, early and accurate detection is essential. Due to their heavy reliance on clinical knowledge, traditional diagnostic techniques are prone to subjectivity. In order to overcome these obstacles, automated skin cancer diagnosis has been using machine learning (ML) and deep learning (DL) models more and more. Convolutional Neural Networks (CNNs), Vision Transformers (ViTs), and ensemble learning models are among the ML and DL models that are methodically assessed and contrasted in this study for their ability to classify skin lesions. We examine how classification performance is affected by preprocessing methods, optimization tactics, and dataset selection. More specifically, this study makes use of publically accessible benchmark datasets including PH2, ISIC, and HAM10000 to guarantee a thorough assessment of model effectiveness. Our results show the benefits and drawbacks of various approaches, offering guidance for creating AI-driven diagnostic tools that are more precise, understandable, and useful for actual clinical settings.