Machine learning predicts housing value more accurately
Marnix Hazelhoff
26 September 2025
Reading time 3 minutes
Folmer, E., & Kuffer, M. (2022). Spatial Determinants of Real Estate Appraisals in The Netherlands: A Machine Learning Approach. ISPRS International Journal of Geo-Information, 11(2), 125.
The housing market is complex, and no home or neighborhood is the same. Yet property valuations are often based on relatively simple models. New research shows that modern machine learning techniques are much better at capturing this complexity. The role of location in particular proves to be crucial.
Why this research?
Buyers, sellers, and lenders all rely on property valuations. Traditional models often use linear relationships: more square meters means a higher price, and newer homes are generally worth more. But that picture is too simplistic. A home’s value is also strongly influenced by its surroundings: the quality of the neighborhood, proximity to amenities, and even subtle differences in housing type.
​
To investigate whether better value estimates are possible, Dutch researchers applied machine learning. This method can detect patterns that traditional models miss.
Method
The researchers used housing and appraisal data from several municipalities and compared two approaches:
-
Classical regression models: the standard in the real estate world.
-
XGBoost: an advanced machine learning technique that can capture both linear and complex, non-linear relationships.
​
The models were fed with a wide range of data, including:
-
Housing characteristics (floor area, type, year of construction, WOZ-value)
-
Environmental characteristics (amenities, neighborhood features)
This allowed the researchers to assess which factors play the largest role and how well each model could predict the appraised value.
Key findings
-
Higher accuracy: XGBoost explained about 83% of the variation in housing values, far more than classical regression models.
-
Location is decisive: a home’s value is not only determined by what is inside its walls, but just as much by the neighborhood and nearby amenities.
-
Transparency remains difficult: machine learning predicts better, but it is not always clear to users why a certain value comes out of the model. This can lead to mistrust.
Calculate bid
Do you want to know what to bid on your dream home? We calculate it for you.
What does it mean for the housing market?
This study shows that the real estate sector can achieve more reliable valuations with machine learning. But the authors also warn: if the logic behind predictions is not transparent, this can cause problems for appraisers, banks, and consumers. Accuracy must therefore always go hand in hand with explainability.
