CODEWITHSUNDEEP
machine-learningcluster-analysispattern-recognition
Elkan
Elkan

Reputation: 616

How can we interpret negative adjusted rand index?

Adjusted rand index (ARI) is a popular measure to compare two clusters. Unfortunately, I usually get negative ARI after performing clustering analysis and comparing them. How can I interpret these negative ARIs to describe the differences of those clusters? And then if the negative ARIs are meaningless, any suggestion about an appropriate measure?

Upvotes: 4

Views: 4928

Answers (2)

coulminer
coulminer

Reputation: 368

I recently had a similar interpretation question and found these toy examples in R useful:

> aricode::ARI(c(1,1,2,2), c(1,1,2,2))
[1] 1
> aricode::ARI(c(1,1,2,2), c(2,2,1,1))
[1] 1
> aricode::ARI(c(1,1,2,2), c(1,2,1,2))
[1] -0.5
> aricode::ARI(c(1,1,2,1), c(1,2,1,2))
[1] 0

Upvotes: 0

Has QUIT--Anony-Mousse
Has QUIT--Anony-Mousse

Reputation: 77485

They aren't "meaningless" at all.

Negative ARI says that the agreement is less than what is expected from a random result. This means the results are 'orthogonal' or 'complementary' to some extend.

But this shouldn't happen often, unless you deliberately look for alternative clusterings. Maybe there is an implementation error?

Upvotes: 8

Related Questions