Building Vacancy Detection With
GeoAI
Using Proxy Data in a Constrained and Mixed Access Environment
Darragh Redmond, Milo Dennehy, Junli Liang, Katsiaryna Bahamazava, Rob Brennan, Gavin McArdle ABSTRACT
This article presents a GeoAI model designed to predict building vacancy in Ireland using only open and licensed‐but‐accessible datasets. It operates under mixed‐access, proxy‐based conditions, as legal and technical barriers frequently restrict access to proprietary data like utility consumption. Our model achieved a 0.379 F1‐score and 0.348 F1‐score under temporal holdout conditions, establishing a new baseline, with performance proving stronger in urban areas compared to rural regions. These results suggest limitations when relying solely on the available proxy data, highlighting the difficulty of detecting vacancy in environments where utility data are not readily available. This study prioritizes a schema‐agnostic, reproducible architecture and governance framework over model complexity to handle the realities of data scarcity and interoperability in urban planning.