AI decodes the architectural DNA of Singapore's shophouses, NUS study finds

Machine learning has turned Singapore's shophouses into a data problem — and the results are rewriting how the country understands its own built heritage. Researchers at the National University of Singapore used AI and mathematical techniques borrowed from evolutionary biology to reconstruct how the island's iconic shophouses developed as a connected architectural "family tree," in a study published in Nature Communications that analysed 1,276 shophouses in Chinatown.

Led by Professor Heng Chye Kiang of NUS's Department of Architecture, the research created a repeatable computational framework for studying vernacular architecture — buildings shaped by local practice, available materials and cultural exchange rather than formal design plans. The team assembled georeferenced images of shophouse frontages from Google Street View and field photography, used an AI object-detection model to identify 14 architectural elements across the façades, and converted each building into a structured profile of features present or absent.

The striking step came next: the researchers applied phylogenetic analysis, a family of methods traditionally used to trace relationships among biological species, treating shared façade features as architectural traits. The resulting network organised Chinatown's shophouses into nine distinct clusters — and rather than a single chronological ladder of styles, it showed several architectural tendencies existing at the same time, with some styles evolving locally while others diffused between communities. Forms associated with different ethnic groups coexisted in parallel lines of development rather than blending into one uniform style.

The findings challenge the expectation that architectural history must be a single orderly progression. In a multi-ethnic trading city, a façade element can be evidence of social connection, cultural competition or adaptation to local circumstances. The researchers argue the framework could change how historic buildings are documented and conserved — a quantitative genealogy showing how individual façades relate to broader patterns, and which buildings occupy important positions in the network even if they look visually unusual.

Why it matters for Singapore: Heritage conservation decisions — what to preserve, how shophouse districts evolve — have traditionally rested on expert judgment. This study shows AI can scale that judgment across thousands of buildings and surface patterns invisible to the eye, giving conservation authorities a new evidence base. It is also a small but telling example of what Singapore's AI push means beyond the economy: the same tools driving exports and productivity are now being turned on the country's own history and identity.