The building sector, which contributes to nearly one-third of global carbon emissions, is crucial for achieving carbon neutrality worldwide. The non-domestic building stock (NDBS) is an energy-intensive subset, and its long lifecycles make it essential to analyse it to help achieve carbon neutrality targets. Until now, the heterogeneity of the non-domestic building stock, coupled with the lack of publicly accessible data have created significant barriers to the creation of high-resolution energy models. Therefore, there is a need to explore some reliable means that can be used to interpolate building stock databases. In this study, using hotel buildings in England and Wales as a case study, public data such as Energy Performance Certificates, Valuation Office Agency rating lists and UK Buildings are integrated, and deep learning means are employed to interpolate the age-of-construction attributes of the buildings to refer to contemporaneous legislation to obtain hypothetical values of the building envelopes for subsequent stock level analyses. The study demonstrates that this means is stronger than the traditional means of randomly assigning attributes to buildings based on the overall distribution of the stock, with higher accuracy (>90%), and can be used to build a reasonable database for constructing physics-based models of building stock. The method is generalisable and can potentially be used to impute missing attributes of NDBS in other countries with similar databases.