A novel housing price estimation model integrating GIS and DNN: a case study of Istanbul

Abstract

Housing valuation is a concrete reflection of socio-economic inequalities in urban space. Particularly in densely populated, spatially fragmented large cities like Istanbul, the current official mass appraisal system used for property taxation fails to reflect market reality. This situation results in revenue losses in property taxes and spatial injustices. This study develops a Deep Neural Network (DNN)-based model that integrates spatially derived variables from Geographic Information Systems (GIS) and incorporates 24 objective variables related to location, structure, and access in Istanbul. The model, trained on 3,757 samples created using open-source big data, estimated housing values with high accuracy ($R^2$ = 0.979). The findings show that spatial differences in housing values are strongly related to urban variables such as accessibility and proximity to infrastructure. This approach not only produces housing value estimates but also provides a theoretical and methodological framework for spatial analyses of how value is produced in urban space. The study has the potential to support the development of a fair, transparent, and updatable mass appraisal system, especially for developing cities.