使用r中的邻近矩阵进行聚类

Question

I have a proximity matrix (dissimilarity) of an mahalanobis distance.

the matrix (sample):

> dput(MD[1:5,1:5])

structure(c(0, 10.277, 8.552, 8.592, 9.059, 10.277, 0, 10.917, 
9.489, 8.176, 8.552, 10.917, 0, 8.491, 8.104, 8.592, 9.489, 8.491, 
0, 9.375, 9.059, 8.176, 8.104, 9.375, 0), .Dim = c(5L, 5L), .Dimnames = list(
    c("2", "4", "5", "6", "9"), c("X2", "X4", "X5", "X6", "X9"
    )))

the matrix has 1900 people and the row name are an Id. I need to cluster those people and the to get a number of a cluster next to the person's id.

I know how to cluster using k-means but I don't how to cluster when you have already a dissimilarity matrix.

Answer 1

You can use hierarchical clustering, starting with the Mahalanobis distance matrix:

MD
      X2     X4     X5    X6    X9
#2  0.000 10.277  8.552 8.592 9.059
#4 10.277  0.000 10.917 9.489 8.176
#5  8.552 10.917  0.000 8.491 8.104
#6  8.592  9.489  8.491 0.000 9.375
#9  9.059  8.176  8.104 9.375 0.000

hc <- hclust(as.dist(MD))

clusters <- cutree(hc, k = 3) # obtain 3 clusters
clusters
#2 4 5 6 9 
#1 2 3 1 3 

plot(hc)
rect.hclust(hc, k = 3, border = "red")

Answer 2

If I understood you correctly, using dendextend :

fit<-hclust(MD)

fit %>% as.dendrogram %>% 
  set("branches_k_color", k = 3, value = c("purple", "orange","red")) %>% 
  plot

The cluster # are specified by colors ... passed on from cutree .