How to group scattered points that visually belong to lines?

Hi there,

I have a set of scattered points look like this:

The data are stored in a x-vector and a y-vector and the sequence is random.

By visual, it is apparent that there are five lines from left to right and a V-shaped group of points, but how can I group the points that fall on the lines into separate arrays by code?

Thanks.

[Adam Danz]

I wonder if kmeans clustering would work.

https://www.mathworks.com/help/stats/kmeans.html

[Akira Agata]

Or "Hierarchical Clustering" would work.

https://jp.mathworks.com/help/stats/hierarchical-clustering.html

If possible, I (... and many experts here) would be happy if you could upload your data here.

[Hong-Cin Liou]

Hi Adam and Akira,

Thanks for your suggestions! Here I provide my data file if this can help a bit. The data has two vectors 'X' and 'Y'.

Answer 1

[Adam Danz]

The data you provided was helpful. This should get you started. Check out the documentation for clusterdata() so you can tweak this if you decide to continue with this route.

The 'idx' variable below indicates the group number for each (X,Y) pair.

load('scattered points.mat')
idx = clusterdata([X',Y'],6); 
gscatter(X,Y,idx)

[Hong-Cin Liou]

Thank you Adam, this is really helpful!

One follow-up question: do you think it's possible to automatically detect the number of groups? I have tried to rotate the axes and looked at the projection on one of the axes. Although this did work, I am still wonderin if there's better way...

[Adam Danz]

Thanks, glad I could help. There are a few ways to estimate the number of clusters but this is a common problem without a one-size-fits-all solution.

Search for "the elbow method of clustering" for one common approach.