Abstract:
The improvement of data management and data capturing techniques has led to the
availability of large amounts of data for analysis. This is especially true in the field of spatial data,
where data is indexed by location. Traditionally, spatially correlated data has been analysed using
methods that rely on the spatial component of the data. This article will compare the results of
using traditional spatial methods such as Kriging and geographically weighted regression against
the use of other statistical data mining methods, given the large amount of data available. Using a
dataset containing property values for the Tshwane Metropolitan area, different spatial and statistical
models will be applied for predictive purposes in order to determine which model represents the data
most accurately. Finally, these methods will be combined using stacking, to determine whether the
combination of models has better predictive abilities than the single models.
