-2

I am new to python and machine learning. I am trying to cluster my dataset by using the DBSCAN algorithm. But I am stuck with getting correct values for MinPts and eps. I checked several solutions and didn't find a way to choose the values for those parameters. How can I choose those value, better if can explain it briefly.

Andreas Rossberg
  • 34,518
  • 3
  • 61
  • 72
Zeal
  • 19
  • 5
  • 1
    Does this answer your question? [Parameter estimation in DBSCAN](https://stackoverflow.com/questions/15050389/parameter-estimation-in-dbscan) – Simon Zyx May 23 '20 at 12:42
  • Completely I don't have an idea about how many clusters(domain knowledge) should be there as I used location data to cluster the area. – Zeal May 23 '20 at 13:42
  • As stated in the above answer, there is no way to determine good values per se. You have to experiment with different values. Choose different eps and MinPts and see what happens. Do you know if the distances are more likely to be kilometers or nanometers? You can adjust your eps accordingly. Try to visualize your results. Are the points clustered to much? Try reducing minpoints and/or eps. If they are clustered to little try the other way around. – Simon Zyx May 27 '20 at 08:15

1 Answers1

0

In the original publication (section 4.2) of DBSCAN the authors proposed a way to determine good values for MinPoints and eps.

They also ran tests that show that you can elimiate the MinPoints parameter for a 2-dimensional dataset by always using MinPoints = 4. Because there results for values greater than 5 are not significantly different than the ones with MinPoints = 4 but they are computationaly more expensive.

Maxipet
  • 1
  • 1