In our group we are dealing with misplacement of items and we came out with a method derived from knn, but we are not sure if there is already some place where this was described. The method would be this one:

Let us imagine the situation where we have a group of items with different classes, lets say "red", "blue" and "green", situated in a Cartesian space. These items are located in some way closer one to another if they belong to the same class.

Now we introduce a new item with one of the three classes, lets say "red". Then we get the e.g. 10 nearest neighbors in the space and we get 6 red, 2 green and 2 blue. We would then assume this item has a score of 0.6 so probably its location is correct. However, if we had a red item with 2 red, 4 green and 4 blue this item would be probably misplaced since its score is 0.2

We have been looking for references since this is not exactly knn clustering. Does anybody of you know if this approach was used before? If so, does the similarity measure have a name?

Thank you very much! I would also appreciate some input


There is no kNN clustering. There is k-means, but that is unrelated.

What you are doing is kNN classification to check your labels for plausibility. Still, it's simple kNN classification by which you predict the expected label.


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.