I am studying multi-label learning methods, where for a given observation, you can assign more than one (a set of) target labels. One example is multi-label k-Nearest Neighbor.

I am seeking a way to describe to a panel of people unfamiliar with multi-label learning methods, a way to visualize how they work. For multi-label kNN, I need a visualization, much like the single-label multi-label approach found here: How to plot decision boundary of a k-nearest neighbor classifier from Elements of Statistical Learning?.

Note: This is not a duplicate of the above question (that I linked to above), because this is a multi-label version of kNN. The single-label solution is an intuitive visualization process, but the multi-label version is giving me trouble.

Can anyone help me understand how to visualize the predictions from a multi-label k-Nearest Neighbor classifier?

  • $\begingroup$ Like this? stats.stackexchange.com/questions/21572/… $\endgroup$
    – Preston
    Commented Sep 24, 2015 at 21:59
  • 1
    $\begingroup$ Since this refers to the multi-label case, it is arguably not a duplicate, as is explained in the question text. $\endgroup$
    – Silverfish
    Commented Sep 24, 2015 at 22:55
  • 1
    $\begingroup$ Exactly, this is not a duplicate @Preston. $\endgroup$ Commented Sep 25, 2015 at 0:34

1 Answer 1


This is the best visualization I can attempt to use to describe multi-label KNN. Let me know if you disagree.

In the plot below, individuals are one or more of the labels: {blue, orange, green}. As you can see, some individuals are both blue and orange, some green and orange. For the test subject I point to with the red arrow, the 7 nearest neighbors are probed.

Plot in principle component input space, of a three class classification problem, using 7-Nearest Neighbors.

From examination of those 7 nearest neighbors, you get the histogram below, yielding a final ranking class order of: Blue=Orange > Green, meaning this test subject is blue or orange before it is green. I don't know how precisely this translates to class probabilities. Would love to learn more?

Histogram of test subject's class membership based on 7 nearest neighbor's class membership.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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