Abstract
Skin cancer remains one of the most common malignancies worldwide, with early detection and personalised risk stratification being key to improving patient outcomes. In recent years, machine learning (ML) has emerged as a powerful tool in dermatology, offering enhanced diagnostic accuracy, risk prediction, and clinical decision-making as a support for clinicians and patients. This chapter investigates the current and future potential of ML models in the early diagnosis and prognostic modelling of pigmented skin cancers, including melanoma, basal cell carcinoma (BCC), and squamous cell carcinoma (SCC). It examines the evolution of ML architectures, from classical algorithms like support vector machines and random forests to advanced convolutional neural networks (CNNs) and multi-module approaches. Key public datasets such as ISIC and HAM10000 are reviewed alongside their limitations, including the underrepresentation of Fitzpatrick skin types IV–VI. Through comprehensive tables and figures, we synthesise the performance metrics, learning methods, and datasets of leading diagnostic models. Furthermore, we highlight the integration of genomic data, mobile apps, FDA-cleared artificial intelligence (AI) devices, and real-time histopathology tools, as well as the ethical, technological, and regulatory challenges of clinical implementation. Data bias, privacy, and the need for inclusive design to ensure equity across diverse populations will also be discussed. The chapter concludes with a forward-looking perspective on federated learning, augmented reality (AR), and teledermatology (TD), aiming to shape an inclusive and AI-augmented future for skin cancer care.

This publication has 61 references indexed in Scilit: