Difference between DBSCAN and HDBSCAN - NISHIO Hirokazu's Scrapbox (Auto-translated from Japanese)

Difference between DBSCAN and HDBSCAN

Difference between DBSCAN and [HDBSCAN

Simply put, HDBSCAN is a system that automatically adjusts the value of DBSCAN eps

Experiment with real data to observe behavior

https://gyazo.com/b33420c5a98cd016a3eebd26a505ddbe https://gyazo.com/656733acc1eb23cd794e535f551d0786

https://pberba.github.io/stats/2020/01/17/hdbscan/

Illustration of eom(Excess of Mass), the default cluster selection criteria for HDBSCAN

2024-11-14

https://gyazo.com/ff7c969a58a1b63a88fc375355dab761

lower left

HDBSCAN recognizes the lower left "clearly separated cluster" as a whole cluster, regardless of the parameters.

DBSCAN ignores the end part as noise, gradually decreasing in size as the parameters change, and finally judging all of it as noise.

right

There's not much difference in behavior, but HDBSCAN's are more likely to judge the surrounding noise as part of the cluster and get involved.

DBSCAN — scikit-learn 1.5.2 documentation

DBSCAN Revisited, Revisited: Why and How You Should (Still) Use DBSCAN

HDBSCAN — scikit-learn 1.5.2 documentation

https://note.com/navy_azalea/n/na859d7ab6ab3

Mutual reach

SpectralClustering — scikit-learn 1.5.2 documentation

---

This page is auto-translated from /nishio/DBSCANとHDBSCANの違い using DeepL. If you looks something interesting but the auto-translated English is not good enough to understand it, feel free to let me know at @nishio_en. I'm very happy to spread my thought to non-Japanese readers.