1

Background:

So I had a thought, tried it out, and liked what it did. I'm sure someone else has done this. It feels very convenient. It also gives an interesting take on robust nonparametric density estimation.

Process:

  1. copy the dataset so I'm not deleting all the data
  2. while the number of elements remaining in the data is above target percentage
  • Find the convex hull of the data
  • remove those points from the data
  • calculate the number of remaining points
  1. At the target level of remaining points, determine and plot the convex hull

Question:
Can you tell me what the actual process is called, point to some papers, and possibly criticisms showing its weaknesses?

EngrStudent
  • 8,232
  • 2
  • 29
  • 82
  • 1
    https://www.google.com/search?client=firefox-b-1-d&q=convex+hull+peeling. On this site see https://stats.stackexchange.com/questions/41578 (bagplots) and https://stats.stackexchange.com/questions/465054 (Tukey depth). Your method is described in an answer at https://stats.stackexchange.com/questions/221052. I found these by searching for "convex hull outlier". – whuber Dec 08 '20 at 16:50
  • @whuber - would you mind putting that into an answer. – EngrStudent Dec 08 '20 at 16:53
  • 1
    @whuber - it seems there are multiple ways to "skin" that cat. One could find the centroid of the hull and them remove the most distant point. One could do chull peeling in layers. Chasing down references gave me this, which is also quite neat. https://arxiv.org/pdf/1805.08278.pdf – EngrStudent Dec 08 '20 at 18:14

0 Answers0