How to Reassign Wrong cluster Values.

Hello, I hope you are doing well. I have the dataset,Which is wrongly clusters into 4 cluster, but there should be 3 Clusters.

The cluster 2 (cell 2) and cluster 3 (cell 3) have almost similar values. I want to rearrange the clusters to make 3 clusters

How can i do that in Matlab. Can anybody help me with that?

Answer 1

[Image Analyst]

How are you clustering it? If you use kmeans you can tell it (force it) to use 4 clusters and it will find 4 clusters.

[Med Future]

@Image Analyst Yes, i am using K-means by the following command, its says optimal clusters are 4, but my groundtruth have 3 clusters

eva = evalclusters(dataset,'kmeans','silhouette','KList',[1:4]);

[Med Future]

@Image Analyst Can you please comment on this?

[Image Analyst]

I read in your data and found this:

s = load('Dataset.mat')

s = struct with fields:
    clusters: {4×1 cell}

ca = s.clusters

ca = 4×1 cell array
    {9469×4 double}
    {1500×4 double}
    { 630×4 double}
    { 323×4 double}

plotColors = lines(numel(ca));
% Plot first 3 columns
subplot(2, 1, 1);
for k = 1 : numel(ca)
    thisCluster = ca{k}
    plot3(thisCluster(:, 1), thisCluster(:, 2), thisCluster(:, 3), '.', 'Color',plotColors(k, :))
    hold on
end

thisCluster = 9469×4

1.0e+06 * 0.5716 0.0000 0.0000 0.0000 0.5717 0.0000 0.0000 0.0000 0.5717 0.0000 0.0000 0.0000 0.5718 0.0000 0.0000 0.0000 0.5719 0.0000 0.0000 0.0000 0.5719 0.0000 0.0000 0.0000 0.5719 0.0000 0.0000 0.0000 0.5719 0.0000 0.0000 0.0000 0.5720 0.0000 0.0000 0.0000 0.5720 0.0000 0.0000 0.0000

thisCluster = 1500×4

1.0e+06 * 0.1716 0.0001 0.0000 0.0000 0.1746 0.0001 0.0000 0.0000 0.1855 0.0001 0.0000 0.0000 0.1888 0.0001 0.0000 0.0000 0.2012 0.0001 0.0000 0.0000 0.2046 0.0001 0.0000 0.0000 0.2052 0.0001 0.0000 0.0000 0.2063 0.0001 0.0000 0.0000 0.2069 0.0001 0.0000 0.0000 0.2074 0.0001 0.0000 0.0000

thisCluster = 630×4

1.0e+06 * 0.1719 0.0001 0.0000 0.0000 0.1743 0.0001 0.0000 0.0000 0.1749 0.0001 0.0000 0.0000 0.1845 0.0001 0.0000 0.0000 0.1871 0.0001 0.0000 0.0000 0.1993 0.0001 0.0000 0.0000 0.2038 0.0001 0.0000 0.0000 0.2050 0.0001 0.0000 0.0000 0.2061 0.0001 0.0000 0.0000 0.2066 0.0001 0.0000 0.0000

thisCluster = 323×4

1.0e+06 * 1.4845 0.0000 0.0000 0.0000 1.4848 0.0000 0.0000 0.0000 1.4855 0.0000 0.0000 0.0000 1.5088 0.0000 0.0000 0.0000 1.5095 0.0000 0.0000 0.0000 1.5099 0.0000 0.0000 0.0000 1.5102 0.0000 0.0000 0.0000 1.5126 0.0000 0.0000 0.0000 1.5129 0.0000 0.0000 0.0000 1.5136 0.0000 0.0000 0.0000

xlabel('Column1')
ylabel('Column2')
zlabel('Column3')
grid on
% Plot last 3 columns
subplot(2, 1, 2);
for k = 1 : numel(ca)
    thisCluster = ca{k};
    plot3(thisCluster(:, 2), thisCluster(:, 3), thisCluster(:, 4), '.', 'Color',plotColors(k, :))
    hold on
end
xlabel('Column2')
ylabel('Column3')
zlabel('Column4')
grid on

So it seems reasonable that there could be 4 clusters. Can you tell me what the 4 columns mean? Is it possible one of them should not be used in the decision as to what class a row belongs in?

[Med Future]

@Image Analyst The all four columns are the features which are used to identify the cluster.

I used column 2 and column 4 for K-means clustering.

[Image Analyst]

Your data called "clusters" has 4 datasets, so why do you say that there should be 3? When you run it, it looks like 4 clusters is reasonable. Which sets should be combined? And when you told kmeans there were 3, which sets did it combine?

s = load('Dataset.mat')

s = struct with fields:
    clusters: {4×1 cell}

ca = s.clusters;
plotColors = lines(numel(ca));
% Plot first 3 columns
for k = 1 : numel(ca)
    thisCluster = ca{k};
    plot(thisCluster(:, 2), thisCluster(:, 4), '.', 'Color', plotColors(k, :), 'MarkerSize', 40)
    hold on
end
xlabel('Column2')
ylabel('Column4')
grid on

Answer 2

[Jiri Hajek]

Hi, your clusters contain just matrices organized in four columns, so you can easily e.g. put two of them together (concatenate versitcally) like this:

clusters{3} = [clusters{3};clusters{4}];

And to remove the fourth cluster, you can use:

clusters = clusters(1:3);

[Med Future]

@Jiri Hajek But i want it to be using any algorithm, not like merging into one another

[Jiri Hajek]

Since you did not mention any specific algorithm, this was actually a valid response.:-) But jokes aside, try to explain what you need. You want to rearrange but not to merge - so perhaps you should try to explain wat you want to achieve with the data, exactly.

[Med Future]

@Jiri Hajek Let me explain this to you, I have apply clustering algorithm on this, There should be 3 Clusters, but the clustering algorithm solve this into 4 clusters

The cell 2 and cell 3 are mostly similar values, we want to apply any algorithm which have similar values or values near to each other should be in 1 cell.