apply knn over kmeans clustering

In indoor localization if i did clustering a dataset by k-means clustering algorithm then is it possible to apply knn algorithm on that clusters to find positioning error.

Answer 1

[Image Analyst]

No, I don't think so. kmeans() assigns a class to every point with no guidance at all. knn assigns a class based on a reference set that you pass it. What would you pass in for the reference set? The same set you used for kmeans()? If so, then there would be no error whatsoever because knn would simply assign each point to the same class that kmeans() did!

You would need to explain this better so that we know your thought process.

[Supriya Gain]

Actually I'm doing indoor localization and for that i'm using kmeans clustering algorithm. Now i want to use knn algorithm as positioning technique. is it possible?

[Image Analyst]

What would your data be nearest TO? You need a ground truth reference set so it can compare the distances of your test data to every point in the reference set. If you have that, then yes, you can do it.

[Supriya Gain]

I don't understand what 'TO' means. I have a dataset which consist of location coordinates and in a specific location co-ordinate i have signal strengths of 9 wifi routers.

[Image Analyst]

So you have N data points which consist of only their (x,y) locations. So you have a N by 2 matrix of (x,y) locations.

And you have another (x,y) location - a single location - let's call it (xRef, yRef) where you have a vector with 9 values that represent the signal strengths, at (xRef, yRef), of 9 routers that located somewhere else.

Is that the situation? Where are the routers located? Does that matter?

OK, so what are your classes? Are the classes the cluster number that the N points fall into? Are the classes which router (not sure how that could work)?

[Supriya Gain]

No. not like that

I'm explaining you. I have a excel data sheet. In that sheet in a single row A1,B1 is X,Y co-ordinate and C1 to K1 i have signal strength from 9 wifi routers. Now i have 80 rows like that.

And routers location doesn't matter.

Hope you understand now.

[Image Analyst]

OK, but I still don't know what your classes represent. How many classes did you say you have? Like 3 for low strength, medium strength, and high strength router signal at the test point? Attach your workbook if you want us to use it.

For what it's worth, I'm attaching a demo where I determine the location of an explosion based on the value (time) at certain points. Kind of like if you wanted to locate each router based on the signal strengths at each of the 80 points.