Performance evaluation for kmedoids clustering?

Hi, how can i evaluate performance of kmedoids clustering because the available function evalclusters such as silhouette, gap and etc only accept kmeans, linkage and gmdistribution as valid value for clust argument. Anyone can share any file exchange for kmedoids clustering evaluation? Thank you

Answer 1

[Mann Baidi]

Hi @husnir nasyuha,

As per the understanding of the question you would like to evaluate the performance of the kmedoids clustering algorithm for different cluster numbers in MATLAB.

I would like to inform you that you can use the 'evalclusters' for evaluting the kmedoids clustering algorithm also. You can evaulate custom clustering algorithm using the 'evalclusters' function.

You can take the help of the following code example for evaluating your data:

% Generate some sample data
 clear;
data = 100*rand(1000, 2)+ 10*rand(1,1)+ 2*rand(1,1); % 100 samples, 2 features

% Set the number of clusters (K)
K = 10;
cluster_label = zeros(size(data,1),K);
for i=1:K
cluster_label(:,i) = kmedoids(data, i);
end
eva = evalclusters(data,cluster_label,'DaviesBouldin')

eva = 
  DaviesBouldinEvaluation with properties:

    NumObservations: 1000
         InspectedK: [1 2 3 4 5 6 7 8 9 10]
    CriterionValues: [NaN 1.1532 0.8556 0.7692 0.8740 0.8346 0.7993 0.8218 0.7996 0.8407]
           OptimalK: 4

plot(eva)

For exploring more about evaluating the cluster solution using the 'evalclusters', you can refer to the example mentioned in the following link:

https://www.mathworks.com/help/stats/evalclusters.html#bt0_87h-1

I hope this will help in resolving your issue!