How to decide on the correct number of clusters?

Determining the number of clusters/segments in hierarchical clustering/segmentation algorithms

 

由于unique函数在这里找矩阵不同元素的时不能有效得到，所以自己编了小程序实现。

 

Matlab codeL:

clear
load fisheriris
figure(1)
Z = linkage(meas,'ward','euclidean');%Create a hierarchical binary cluster tree using linkage
[Hr,Tr]=dendrogram(Z);  %generates a dendrogram plot of the hierarchical binary cluster tree

val0=eval(vpa(Z(:,3),5));

t=1; n(1)=0; val=val0(1); f(1)=val;
for i=1:length(val0)
    if val0(i)==val
        n(t)=n(t)+1;
        continue
    else
        t=t+1;
        val=val0(i);
        n(t)=1;
        f(t)=val;
        continue
    end
end
figure(2)
fsum=cumsum(n);
plot(length(val0)-fsum, f, 'o-','LineWidth', 2)
xlabel('Number of Clusters')
ylabel('Merge Distance')
title('A sample evalution graph')