Housing price variations using spatio-temporal data mining techniques

The issue of property evaluation and appraisal has been of high interest for private and public agents involved in the housing industry for the purposes of trade, insurance and tax. This paper aims to investigate how different factors related to the location of a property affect its price over time. The predictive models applied in this research are driven by real estate transactions data of Tehran Metropolitan Area, captured from open data available to the public. The parameters of the functions that describe the behavior of the housing market are estimated through applying different types of statistical models, including ordinary least squares (OLS), geographically weighted regression (GWR) and geographically and temporally weighted regression (GTWR). This suite of models has been run in order to compare their efficiency and accuracy in predicting the variations in housing price. The GTWR model showed significantly better performance than OLS and GWR models, as the goodness of fit index (adjusted R²) improved by 22 percent. Therefore, spatio-temporal non-stationary modelling is significant in the explanation of the variations in housing value and the GTWR coefficients were found more reliable. Three internal factors (size of building; building age; building quality), and eight external factors (topography; land-use mix; population density; distance to city center; distance to subway station; distance to regional parks; distance to highway; distance to airport) influence the property price, either positively or negatively. Moreover, using significant variables that extracted from regression models, the optimum number of housing value clusters is generated using the spatial ‘k’luster analysis by tree edge removal (SKATER) method. Five clusters of housing patterns were recognized. The policy implication of this paper is grouping of Metropolitan Tehran housing value data into five clusters with different characteristics. The varying factors influencing housing value in each cluster are different, making this data analysis technique useful for policy-makers in the housing sector.