2
$\begingroup$

I am new in datamining. I decided to play with k-means clustering in r with simple аrtificial data.

set.seed(101)    
x1 <- runif(100,36.6,37.5)
x2 <- runif(100,37.6,38.4)
x3 <- runif(100,38.5,40)
x<-c(50,150,250)
y<-c(37,38,39.25)

centers<-c(37.05,38,39.25)
all <- c(x1,x2,x3)
c1<-kmeans(all,centers=centers,iter.max = 1000,nstart=1,algorithm="Lloyd")
plot(all,col = c1$cluster)
points(x,y,col="red",pch=19)

I generated three random data sets x1,x2,x3. Then i taked center of each cluster like mean of each set and used k-means algorith of r. In result i get what you see in picture.

What i am making wrong?

Why second cluster contains the part of third set? How to improve result?

Why result is so bad instead sets is lineary separate?

The main question is how to improve clustreing result? enter image description here

$\endgroup$
3
$\begingroup$

A 1 dimensional data set is "linearly separable" at any set of points between the data points. Kmeans isn't a nearest-neighbour type of clustering algorithm, it just divides the space so that each point belongs to its nearest cluster centre, so your data could have huge gaps between the source clusters, but the algorithm would be "wrong" if the clusters had different ranges.

For example, if you generate points from two clusters on the intervals (0,1) and (2,20), even though there's a gap between 1 and 2, the clustering algorithm will still assign a lot of the points from the higher source cluster to the lower cluster because they are nearer that cluster centre. Its distance to cluster centre for each points that counts, not distance to neighbours. Use another clustering algorithm (like nearest-neighbours) if you are expecting gaps between clusters.

| improve this answer | |
$\endgroup$

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.