2

I have location posts from Instagram Data and want to cluster them by density - using DBSCAN. I played around quite a lot with different minPts and eps but never receive clusters (always one big cluster) - even if this is wordwide data. Does anyone have a suggestion on how to choose the parameters in order to receive a suitable result?

Enclosed the file I am playing with my file

  • 1
    what range of values have you tired? – Ian Turton Jan 27 at 16:53
  • Dear Ian, 0.1 to 1000 – Roman Egger Jan 27 at 17:36
  • 3
    the issue is likely due to your point coordinates being Lon/Lats (i.e. a geographic CRS), where values between 0.1 - 1000 loosely corresponds to 10km - 100000km (along the equator)... – ThingumaBob Jan 27 at 18:01
  • So what can I do? Tried now to convert it to a shp-file (that´s what I googled but doesn´t work either) – Roman Egger Jan 28 at 10:14
1

Based on @ThingumaBob's suggestion, it seems the issue is that you're working in a geographic CRS, where the units are degrees. Since each degree has an inconsistent size depending on where you are on the globe, this results in clusters of wildly different sizes. You need to you need to reproject your data into an projected CRS. Given that your data is worldwide and you're trying to use a tool based on distance, I suggest a world equidistant projection. Options include:

  • World_Azimuthal_Equidistant
  • World_Equidistant_Conic
  • World_Equidistant_Cylindrical
  • World_Two_Point_Equidistant

See Wikipedia's List of map projections for a discussion of the different projection types.

Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.