0

I want to make cosine similarity a distance measure for undersampling of fraud data , but the problem is if I get the zero values it will give 0/0, what can I do for this what can be the alternative similarity measures that are also distance measures.

Myra
  • 1
  • 1
  • See http://stats.stackexchange.com/a/36158/3277. _Squared_ euclidean distance is `2(1-s)` where `s` is cosine similarity. – ttnphns Aug 05 '15 at 10:21

1 Answers1

0

What about the following?

$dist = 1 - sim$ where $sim$ denotes the cosine similarity.

NeuroMorphing
  • 525
  • 2
  • 12