45th Congress of the European Regional Science Association: "Land Use and Water Management in a Sustainable Network Society", 23-27 August 2005, Amsterdam, The Netherlands
In this work the idea is to analyse general spatially identifiable housing market related data on Dutch districts (wijken) with the SOM (Kohonen Map) and a GIS. One of the authors has earlier carried out purely visual SOM analysis of that data, where patterns formed on a larger 'map' (the output matrix of the SOM) were used as a basis for classification of the Dutch housing market segments on a nationwide level. This way the SOM was used as a method for exploratory data analysis. Now we attempt a more rigorous method of determining the segmentation using a smaller 'map' size, in order to be able to export the SOM-output directly to a GIS-system to analyse it further. Two technical issues interest us: one, the robustness of the results – do the five basic housing market segments found in the earlier analysis prevail (we call these urban, urban periphery, pseudo-rural, traditional, and low-income segments); and two, which classes fit the real situation better and which worse, when using the RMSE for a measure of goodness? We also keep an eye on policy implications and aim at comparing our classifications with the 'actual' ones used in official discourse.