K-means clustering implementation on physiological signals data

I need your kind guidance in using K-mean for clustering on physiological signals data. I have 20 signals file, each file has 14 features and the size is different for each file such as 27X14, 22X14 etc. Now i want to give all these files at a time as input to K-mean to generate the clustering. Bit it always create wrong scatterplot for clusters, as can be seen from plot, there is large number of data at 0, which looks wrong to me. I zoomed in this data in another plot to show you all. I tried to reshap my data but still cannot create the right cluster. I am attaching the code which i am using and the scatterplot. I also attached the data file. Please guide me how to do this or is there any other code to do. Looking forward to hearing from you. Thanks

   cd 'D:\Research\Classification\final'
   labels1=dir('*.csv');
   figure
   IMF_U = [];
   for j =1: length(labels1)
   sw = readmatrix(labels1(j).name);
   % sw =reshape(sw,[],2);
   IMF_U=[IMF_U;sw];
   end
   IMF_U =reshape(IMF_U,[],2);
   [idx,C] = kmeans(IMF_U,3);
   gscatter(IMF_U(:,1),IMF_U(:,2),idx,'bgm')
   hold on
   plot(C(:,1),C(:,2),'kx')
   hold off
   legend('Cluster 1','Cluster 2','Cluster 3','Centroid')

[Image Analyst]

If you have any more questions, then attach your data and code to read it in with the paperclip icon after you read this:

TUTORIAL: How to ask a question (on Answers) and get a fast answer

[John D'Errico]

You appear to have some information in your data set that is all zero. And of course, kmeans found it. What would you expect? If you want better help, then you need to provide the data. Attach it as @Image Analyst showed.

[Muhammad Hammad Malik]

Thank you for the response. I have attached the data file which i am using. There are values in negative and near to zero's too. Kindly have a look. Thanks

Answer 1

[Image Analyst]

I made some changes but what I can't figure out is why you reshape your data from 14 columns into only 2 columns. Why are you doing that?

% Optional initialization steps
clc;    % Clear the command window.
close all;  % Close all figures (except those of imtool.)
clear;  % Erase all existing variables. Or clearvars if you want.
workspace;  % Make sure the workspace panel is showing.
format long g;
format compact;
fontSize = 18;

%    cd 'D:\Research\Classification\final'
fileList = dir('feat*.csv');
figure('Name', 'kmeans data');
IMF_U = [];
for k = 1 : length(fileList)
    % Get data for this file only.
    thisFilesData = readmatrix(fileList(k).name);
    % sw =reshape(sw,[],2);
    IMF_U = [IMF_U; thisFilesData];
end

% Reshape from 14 columns into 2 columns.  WHY?????
IMF_U =reshape(IMF_U,[],2);

% Specify the number of classes we want to force the data into.
numberOfClasses = 3;
% Do the kmeans clustering.
[classAssignments, classCentroids] = kmeans(IMF_U, numberOfClasses);
% Plot the 2 columns.
gscatter(IMF_U(:,1),IMF_U(:,2), classAssignments, 'bgm', '.', 30)
hold on
grid on;
% Plot class centroids with a big black X.
plot(classCentroids(:,1), classCentroids(:,2),'kx', 'LineWidth', 2, 'MarkerSize', 20)
hold off
legend('In Cluster 1','In Cluster 2','In Cluster 3','Class Centroids')
fprintf('All Done!\n');

Other than that the results look reasonable.

[Muhammad Hammad Malik]

Thanks for the quick answer. I tried your code, it is working. But if you zoom in the plot, you can see lot of data is at zero, i could not figure it out why it is there actually. That is why i also tried to use reshape to convert it into vector and then to look in the plot, but it's not working.

[Image Analyst]

Yeah, because you reshaped your data. You have 14 features so don't try to say there is only 2 by combining totally unrelated features. You're just scrambling your data into garbage. Use the full 14 features. You can still get 3 clusters no matter if there are 2 features, 3 features, 14 features, or any number of features.

[Muhammad Hammad Malik]

Thank you, i got your point, but if even i don't use reshaped, i am still getting lot of points just at 0. Is it because there are many values near to zero or anyother reason? If so, it's normal then, right?

[Image Analyst]

Features (columns) 1, 3, 4, and 12 (especially those) have values close to zero. But that doesn't matter. It can still classify the measurements (rows) into 3 classes. If you look at the class assignments:

>> classAssignments

classAssignments =

1

1

1

1

1

1

1

1

1

1

1

1

3

3

3

3

3

3

1

1

2

you'll see that every row got classified as some class/cluster, and all 3 classes are represented. It's impossible to visualize a 14-dimensional space to see the clusters, but, guaranteed, they are in clusters (because you forced it to find 3 clusters). If you just scatter 2 of those features against each other, you may or may not see clustering. Just because you have clusters when considering all 14 features does NOT mean that you will notice the clusters in 2-dimensional space where you plot only a pair of features against each other. You might or might not see clusters, or it might depend on the two or three features you choose to show in the scatter plot.

Does that make sense?

By the way, you can see the last row is the only one in class 2 because that row looks absolutely nothing like the rows above it, so it's its own class.

There is something different between rows assigned to class 1 and those assigned to class 3 but it's difficult to see exactly what is the main reason because there are 14 features and we can't plot in 14-dimensional space to visualize it.

[Muhammad Hammad Malik]

Alright, thanks alot for detailed explaination, i got your point.