The original isolation forest paper states that the algorithm works best on small subsamples, but is it okay to use it on large sample sizes or are other anomaly detection algorithms better?

  • $\begingroup$ Could you include resources like the link to the paper you specified? $\endgroup$ – Pluviophile Jun 17 '20 at 6:20
  • $\begingroup$ Depends. What bad thing are you worried might happen if you use isolation forest on a large data set? Does the article you mention have anything that might answer that question? $\endgroup$ – Sycorax Jun 17 '20 at 14:42
  • $\begingroup$ @Pluviophile cs.nju.edu.cn/zhouzh/zhouzh.files/publication/… $\endgroup$ – ddx Jun 17 '20 at 20:43
  • $\begingroup$ @Sycorax says Reinstate Monica, I am just curious if this is bad practice since the original paper mentions that the algorithm works best on small sample sizes. It seems to be working great for me, I am just curious if anyone has found that another algorithm works better. $\endgroup$ – ddx Jun 17 '20 at 20:45
  • $\begingroup$ What I mean is "works best" could mean different things. It could mean "this algorithm doesn't scale well because it has a high computational cost that increases rapidly with data size, so it rapidly becomes infeasible" or it could mean "this algorithm produces bogus results on large data but not on small data." I'm trying to delineate in which sense you mean "best." $\endgroup$ – Sycorax Jun 17 '20 at 20:47

Your Answer

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

Browse other questions tagged or ask your own question.