|
I have 100 different objects with around 2000 features each, without labels, and need to cluster them. These 100 objects will typically cluster into 15-20 distinct, easily separable groups. Typically most clusters will have 3-5 members, and the rest will be single members. Using hierarchical agglomerative clustering, I was able to get pretty good results, and easily assign clusters. I am now trying random forests for an alternative approach, and have created a proximity matrix. I am looking for advice on how to assign clusters from it? Right now I have tried an approach to go through each row and cluster those elements together with prob > p. I found p = .5 was too restrictive, and using p > .4 or p > .35 gave me very decent results. What else can I do with the proximity matrix? What is another approach to extract clusters from the proximity matrix? Thank you |
