1
$\begingroup$

In an academic paper, they talk about using a nearest neighbour algorithm to predict the cluster of a new point. And how the number of nearest neighbours is set to 10 in their example. What do they mean with this? The two things I could think of were: 1. Look which 10 points used in the training set (neighbours) are closest and then assign it to the cluster of which the majority of the points come from. 2. Collect one by one the closest points from the training set until you have 10 points that come from one single cluster. That is the cluster to which the point belongs.

What are the other ways to assign a(n) (existing) cluster to a new point?

$\endgroup$
1
  • $\begingroup$ Can you post a link to the academic paper? $\endgroup$ – Brian Spiering Jun 6 '20 at 14:27
0
$\begingroup$

If the paper didn't elaborate on this, it must mean that they do 1). They look for the 10 closest neighbors of a new point and use majority voting to assign a cluster. On a side note, strange that they use an even number of neighbors, since it may introduce ties.

"What are the other ways to assign a(n) (existing) cluster to a new point?"

There are plenty of ways but it doesn't mean that they are all good! You could randomly assign points to clusters and use the size of the clusters as weight, but that isn't such a great method.

Generally, a nearest-neighbor approach is what is used. Now, it can be done in many ways:

  • KNN as I explained before (with any number of K)
  • Calculate the mean point of each cluster, and assign to the closest cluster
  • For each cluster, calculate the most distant point to the new point, then assign the cluster that has the "least far" distant point
  • And so on.
$\endgroup$

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.