Machine Learning, Architectural Styles and Property Values

Towards Accountability in Machine Learning Applications: A System-Testing Approach

“…As a result, we argue that the standard metrics testing the prediction accuracy of an ML model-such as the F 1 score and the Herfindahl index-are not sufficient to evaluate the model's performance. Lindenthal & Johnson (2021) also discuss this potential misclassification issue, specifically in the context of housing quality and urban environments. They find that the automatically collected street images contain irrelevant information like trees and vehicles for classifying building vintages.…”

Towards Accountability in Machine Learning Applications: A System-testing Approach

“…Utilizing the rich building-level images from Google Street View, Glaeser et al (2018) find that the improvement in building appearance is associated with higher home values in Boston, while the appearance of foreclosed properties depreciates significantly. In the UK, the architectural style is found to be a significant determinant for resale prices, but it has a limited impact on the primary market (Lindenthal & Johnson, 2021). Law et al (2019) show that street image and satellite image data can capture visual urban qualities and improve the estimation of house prices.…”

“…The triumph of ML applications has only started but has revolutionized commerce, personal interactions, entertainment, medicine, government services, state supervision -and research, already (Simester et al, 2020). In real estate and urban studies, a rapidly expanding literature explores the potential of ML algorithms, introducing novel measurements of the physical environments or using these estimates to improve the traditional real estate valuation and urban planning processes (Glaeser et al, 2018;Johnson et al, 2020;Karimi et al, 2019;Lindenthal & Johnson, 2021;Liu et al, 2017;Rossetti et al, 2019;Schmidt & Lindenthal, 2020;Shen & Ross, 2020). These studies, again and again, demonstrate the undisputed power of ML-systems as prediction machines.…”

“…In this paper, we develop systemtesting stages for two ML-enabled use cases: 1) a building vintage classifier and 2) an automatic valuation model (AVM) for residential real estate. Both use cases leverage street-level images of residential real estate as inputs, which are easy to motivate and have been employed in research recently (Law et al, 2019;Lindenthal & Johnson, 2021). In this paper, we do not have an intrinsic interest in the actual predictions these machines produce but focus on model verification tests:…”

“…Next, we augment off-the-shelf image classifiers that have been re-trained to detect architectural styles of English homes (replicating Lindenthal & Johnson, 2021) or the quintile of price regression residuals. 1 Widely available object detection models can reliably locate doors, windows, overall fac ¸ades, trees, and cars, among many other object categories.…”