ABSTRACT This comprehensive study explores the application of advanced machine learning techniques, specifically one‐class autoencoders, for the authentication and attribution of English porcelain artworks. Focusing primarily on the works of William Billingsley (1758–1828), one of England's most celebrated porcelain decorators, we demonstrate how modern computational methods can complement traditional connoisseurship in resolving long‐standing attribution controversies. We present a novel autoencoder architecture trained on five carefully selected handcrafted features—entropy, energy, contrast, homogeneity and Hausdorff‐Besicovitch dimension—that effectively distinguishes genuine Billingsley works from those by other contemporary artists. The methodology is validated through multiple test cases, including authenticated Billingsley pieces from prestigious services (Earl Camden, Prince of Wales, Duke of Northumberland) and works known to be by other artists. Furthermore, we apply this framework to the historically contentious Pendock Barry dessert service (circa 1805–1808), a 90‐piece Derby porcelain service whose decoration has been disputed amongst scholars for over a century. Our experimental results demonstrate classification accuracies exceeding 75% for authentic Billingsley works, with reconstruction error metrics providing quantifiable measures of artistic similarity. This approach offers a complementary tool to traditional expert opinion, adding reliability, objectivity and reproducibility to the field of ceramic art authentication.
Ugail et al. (Mon,) studied this question.