Hey people!
I wrote an article of how we can create a postcode boundary map for a country, in my case Greece.
I would love some feedback on how to make the map more accurate from anyone more experienced!
You can find the article here.
Cheers
This pretty much looks as good as you’ll get with the available data. So probably the first step is to find a way of getting more data in the problematic areas.
You may find that mountains are one of the factors which make you voronoi triangles the wrong shape. Long ago I used voronoi triangles to try & find missing villages in Lesotho, I clipped the original voronoi triangles by a contour derived from SRTM and then looked for villages in the triangles which were over a certain size. There were probably some shards left where the triangle spanned a mountain range leaving a small fragment on the wrong side of the hills. Unfortunately I cant remember the details of the procedure but it certainly removed some of the more glaring discrepancies. I also did it all in QGIS. If I can find the image I’ll post a link here. (The approach bore some similarity to what I did here http://sk53-osm.blogspot.com/2013/07/persistence-in-urban-environment-2.html)).
Another approach you could use is clustering. The Processing plugin from QGIS provides both K-means & DB-Scan clustering.