K-means classification

Question

I have a data set which I know has 3 classes. Say I want to perform k-means classification on this data set, with k=10. I would obtain 10 clusters. How would I merge these 10 clusters into the 3 classes, and based on what rules would I do this (e.g. distance metrics)?

Many thanks

Answer 1

您可以在群集上再次运行k-means，这次设置为k = 3。或者你可以对它们进行期望最大化。或者你可以用暴力破解并测试所有120种方法将10个聚类合并为3个。

正确的做法是从一开始就设置k = 3。具有错误类别的k-means解决方案可能非常不合适。