I have a polygon layer in QGIS containing the parcel data for a city and surroundings, and a point layer with information about properties that have been sold recently. Since there are far fewer points than the total number of parcels, I need to approximate the price of all of my parcels based on what has been sold nearby.

The sought after information in the point layer is the price per square meter, which each one of the points has in its attribute table.

My first idea was using voronoi polygons for this, and running voronoi on the point layer creates a polygon layer which i could join to each parcel afterwards. However, simply using voronoi doesn't produce a very good result since there are a lot of real world factors not taken into account.

Here is an example output from using voronoi: voronoi polygons output

As you can see, there are a couple of concerning things here. Since two adjacent values can differ by a lot, the result can be very unrealistic. There is also no real-world connection to how big each polygon should be, which, combined with my previous point, makes the results even more unpredictable.

I have tried a different method of just getting average price within 5 kilometers or so from each parcel, but this also causes some issues. A fixed search-distance of 5 kilometers works in more remote places where there aren't as many sales, but in more populated areas, a couple of kilometers can make a big difference in what type of area it is, which reflects on the price.

Is there some kind of established method for this or just a better way to do this?

  • Could you use the average of X number of nearest properties, in combination with a distance limit?
    – Matt
    Aug 6 at 15:24
  • @Matt I'm not sure how I would set a distance limit for an arbitrary property. It would have to vary, since a property in the city center doesn't need to take statistics from a point 5km away, but one further out would need to since there may not be a sample close enough.
    – Axekan
    Aug 6 at 16:08
  • That would be handled by using the (10) nearest properties. In a city center they would be within meters rather than kilometers. The distance limit would be for a remote property that does not have (10) neighbours within 5km. And therefore use however many fall within 5km.
    – Matt
    Aug 6 at 16:15
  • 1
    There is a huge literature on predicting house prices, e.g. ijcsma.com/publications/march2019/V7I302.pdf. Relying just to the locations may not give the best result. If you had a larger sample, IDW would be more plausible.
    – fatih_dur
    Aug 7 at 9:05
  • Hedonic House Price Model?
    – Ian Turton
    Aug 7 at 10:21


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.