Machine learning in the documentation and analysis of historical landmark buildings: a systematic review and practical demonstration

Abstract As machine learning (ML) techniques become increasingly integrated into cultural heritage workflows, there is growing interest in their potential to support the documentation, analysis, and conservation of historical architectural landmarks. This study presents a systematic review of peer-reviewed literature from 2017 to 2024 that applies ML to historical landmark buildings, with a focus on architectural classification, structural analysis, restoration, and digital documentation. The review identifies dominant trends, such as the widespread use of convolutional neural networks (CNNs) and the rising application of generative models, as well as notable gaps in regional representation and dataset accessibility. Geographic imbalance remains a significant concern, with heritage sites in Africa, Latin America, and parts of Asia substantially underrepresented in the current ML literature. To complement the review, a lightweight demonstration is presented using a publicly available architectural dataset and a transfer learning approach for style classification. While not a technical innovation, this case study demonstrates how accessible ML tools can support equity-focused, replicable heritage documentation workflows, particularly in underserved or low-resource settings. The findings offer actionable insights for heritage professionals, educators, and researchers seeking to responsibly adopt AI-driven techniques for analyzing and preserving architectural heritage.