Spatial Housing-Price Modelling for Planning Using Hedonic Regression, GWR and Machine Learning

How much is a house worth here, and what would a new light-rail station do to the prices around it? Planners use house prices for affordability, for valuing public investment and for land-value capture. This project compares the classical hedonic regression, geographically weighted regression (GWR) and gradient-boosted trees (LightGBM) on 21,612 real King County (Seattle region) house sales of 2014-2015, with distances to downtown, to light-rail stations, parks and schools from OpenStreetMap. The honest tests are spatial-cluster cross-validation (whole areas held out) and residual Moran's I. The spec's GWR loses to the plain hedonic OLS when whole areas are held out; LightGBM is clearly better than OLS but not clearly better than a GWR tuned inside the training data. A web application values a home with all three models and shows where they disagree. A teaching prototype, not a valuation for lending, tax or compensation.

Category: School of Architecture and Planning

Architecture & Planning Project AP09 from ProTechSC Academy, with graphical abstract, video demonstration and project guidance.