Styjun

I have applied two different clustering algorithms to my data, and I would like to express the commonality among the results of these.

The data is organized as;

• "ID" = Identifier


• "Group_1" = Results from first clustering algorithm


• "Group_2" = Results from second clustering algorithm.

Group_1 is the output of a hierarchical clustering, which had the highest CVI at k = 5, and Group_2 is the output of k-means clustering, which had the highest CVI at k = 10.

I would like to determine the similarity of the results.

Here is the data, which I try to find the similarity of:

structure(list(ID = c(400100L, 400101L, 400106L, 442306L, 443110L, 
443300L, 443301L, 443302L, 443303L, 443304L, 443307L, 443309L, 
443311L, 443312L, 443313L, 443314L, 443316L, 443317L, 443322L, 
443324L, 443328L, 443329L, 443330L, 443331L, 443332L, 443333L, 
443334L, 443339L, 443344L, 443345L, 443351L, 443365L, 443366L, 
443371L, 443378L, 443382L, 443383L, 443388L, 443390L, 443392L, 
443396L, 443398L, 443399L, 443506L, 443507L, 443511L, 443512L, 
443514L, 443521L, 443522L, 443800L, 443802L, 443816L, 443817L, 
443819L, 443820L, 443823L, 443825L, 443828L, 443829L, 443833L, 
443842L, 443855L, 443859L, 443876L, 443877L, 443879L, 444101L, 
444104L, 444202L, 444204L, 444207L, 444251L, 444305L, 444307L, 
444309L, 444312L, 444314L, 444325L, 444327L, 444328L, 444334L, 
444335L, 444339L, 444341L, 444346L, 444359L, 444501L, 444504L, 
444508L, 444509L, 444511L, 444512L, 444514L, 444517L, 444520L, 
444521L, 444547L, 444548L, 444554L, 445101L, 445106L, 445112L, 
445113L, 445115L, 445120L, 445141L, 445302L, 445303L, 445304L, 
445309L, 445312L, 445313L, 445315L, 445316L, 445318L, 445319L, 
445322L, 445327L, 445330L, 445333L, 445404L, 445405L, 445409L, 
445510L, 445522L, 445552L, 445560L, 451704L, 451705L, 452503L, 
452514L), Group_1 = c(1L, 1L, 2L, 2L, 3L, 2L, 4L, 2L, 2L, 1L, 
2L, 2L, 2L, 2L, 1L, 1L, 1L, 2L, 5L, 2L, 2L, 4L, 4L, 4L, 5L, 5L, 
2L, 2L, 1L, 1L, 2L, 2L, 3L, 4L, 4L, 3L, 2L, 2L, 1L, 3L, 1L, 1L, 
3L, 2L, 3L, 2L, 1L, 4L, 2L, 5L, 4L, 5L, 3L, 4L, 1L, 2L, 3L, 2L, 
2L, 5L, 4L, 2L, 2L, 5L, 1L, 1L, 1L, 2L, 5L, 4L, 4L, 2L, 3L, 3L, 
1L, 2L, 1L, 4L, 2L, 4L, 5L, 1L, 4L, 2L, 4L, 2L, 3L, 2L, 2L, 2L, 
1L, 2L, 2L, 3L, 4L, 2L, 2L, 3L, 4L, 1L, 1L, 5L, 2L, 2L, 3L, 4L, 
3L, 5L, 4L, 1L, 1L, 1L, 2L, 4L, 3L, 4L, 4L, 1L, 2L, 1L, 1L, 2L, 
5L, 4L, 4L, 2L, 4L, 3L, 1L, 1L, 3L, 5L), Group_2 = c(7, 7, 7, 
7, 8, 3, 3, 7, 3, 9, 6, 1, 7, 7, 10, 7, 4, 6, 7, 7, 6, 3, 3, 
10, 7, 6, 1, 7, 9, 1, 6, 7, 3, 1, 5, 3, 7, 2, 5, 6, 5, 4, 6, 
10, 1, 1, 1, 10, 1, 6, 7, 6, 6, 3, 7, 7, 6, 5, 7, 6, 9, 7, 8, 
6, 3, 7, 9, 3, 7, 6, 6, 2, 6, 3, 3, 2, 7, 1, 6, 6, 6, 3, 6, 6, 
3, 7, 7, 1, 3, 7, 3, 6, 8, 6, 3, 7, 6, 7, 7, 1, 3, 6, 7, 3, 7, 
3, 7, 3, 3, 5, 5, 2, 6, 3, 1, 6, 7, 6, 7, 5, 2, 7, 6, 5, 7, 1, 
8, 7, 3, 9, 7, 6)), row.names = c(NA, -132L), class = c("data.frame"))

I would like to know a percentage agreement between the two groups, however I cannot figure out how to calculate it.

Ultimately, I would like to arrive at something as:

ID's grouped together in both "Group_1" and "Group_2" divided by N

My assumption would then be that ID's grouped similarly by both algorithms are correctly labelled and I could redo the clustering with the remaining ID's.

Standard clustering evaluation measures such as

• adjusted Rand index (ARI)


• Normalized mutual information (NMI)

can be used to evaluate the similarity of two clusterings. It's easy to see that they are symmetric.