3
$\begingroup$

What are the points called in a dataset where the points have the same features but different labels?

For example, I am trying to predict whether some object is a hot dog. But I only have access to features is_wrapped_in_bread and has_meat_in_the_middle.

| object        | is_wrapped_in_bread | has_meat_in_the_middle | is_hotdog |
| salad         | false               | false                  | false     |
| hotdog        | true                | true                   | true      |
| burger        | true                | true                   | false     |
| veggie burger | true                | false                  | false     |

In the example above, burger and hotdog are not distinguishable based on the available features. Is there a name for that?

$\endgroup$
1
$\begingroup$

I'm not aware of any specific term for this but in general I would call this either:

  • noisy data, typically in the case where these inconsistencies are due to mistakes in the annotation process (normally the proportion of mistakes is much lower than the proportion of correct instances).
  • poor data design, if the way instances/features are defined doesn't match the goal of the ML process (the example provided would fall into this category).
| improve this answer | |
$\endgroup$
  • $\begingroup$ Right, but I want a general term that captures both the notion of potentially bad labels, and potentially bad features. Is there such a term? $\endgroup$ – Daniel Kats Sep 26 '19 at 2:25
  • $\begingroup$ if there is I don't know it. let's see if somebody else has one $\endgroup$ – Erwan Sep 26 '19 at 9:32

Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.