Predicting building age and floor space using feature-engineered 2D urban morphology

Purpose This study aims to develop a generalizable machine learning pipeline that uses only two-dimensional (2D) data to predict building characteristics, specifically construction year and floor space. This addresses the data gaps in building stock models, which are crucial for developing localized greenhouse gas emissions mitigation strategies. Design/methodology/approach Using a novel, national-level building registry dataset from Sweden, we trained machine learning classification models to predict construction year and regression models to predict floor space. The models were developed using only 2D building attributes, avoiding the need for 3D data, which is often unavailable at large geographic scales. Findings The result shows that the best-performing classification model achieves a precision measured as Area Under the Precision-Recall Curve (AUPRC) of 0.823 and the best-performing regression model achieves an R2 of 0.789. These results demonstrate that a 2D-based approach is sufficient for accurately predicting building characteristics. Originality/value This study reveals that imputing missing building attribute data does not require height data. By relying exclusively on widely available 2D data, the proposed machine learning pipeline could overcome the data limitations of previous studies. By demonstrating the effectiveness of this approach on a national-level dataset, this study improves the generalizability of building stock models and provides a scalable solution for estimating building characteristics at larger geographic scales.