Question

I'm trying to cluster points in my dataset. The simple steps are as follows:

  1. Find the nearest neighbor for each point.
  2. Eliminate noise points by setting a threshold for nearest neighbor parameter (those points with large enough nearest neighbor distances are eliminated)
  3. Connect all points that are within a user-specified radius of each other.

I haven't tried this yet but do you think this makes sense? What do you think may be the flaws in this algo? Can we classify it as a form of hierarchical clustering? Is it right that this may be akin to Jarvis-Patrick Clustering and Density-based clustering.

Why is jarvis-patrick not hierarchical? Thanks a lot in advance!

Was it helpful?

Solution

Sounds to me as if you are reinventing DBSCAN.

Your k is called minPts in DBSCAN, and your threshold is epsilon.

OPTICS is a variant that produces a hierarchical result.

Licensed under: CC-BY-SA with attribution
Not affiliated with StackOverflow
scroll top