为什么k均值将观察结果而不是R中的变量聚类？

Question

I have a dataset mydata with 84 variables each with 300 observations, as shown below:

I am using the following code to cluster mydata in 5 clusters:

mydata <- read.csv("mydata.csv", header = TRUE)

# K-Means Clustering with 5 clusters
fit <- kmeans(mydata, 5)

# Cluster Plot against first 2 principal components

# vary parameters for most readable graph
library(cluster) 
clusplot(mydata, fit$cluster, color=TRUE, shade=TRUE, labels=2, lines=0)

This produces the following plot:

I expected it to plot the 84 variables with their names, as shown in the first image, based on the observations. But instead, as can be seen in the last image, it is clustering the 300 observations. How to fix this?

I tried transposing mydata , but that doesn't solve the issue.

EDIT: I expected it to plot something like this (but this plot is for another dataset). I show this plot only to show the names of the variables on the plot, which means the variables are getting plotted (based on observations).

Answer 1

If you want to cluster variables, not instances , you can simply transpose your data matrix.

Usually, clustering is applied to data points, not columns.

Beware of the usual limitations of k-means. It is very sensitive to scale.

Answer 2

The plot you mention had probably been created by using the mtcars dataset:

print(datasets::mtcars)

The points you see are clearly observations.

If you want to create a cluster of variables, multiple options exists:

• Create a matrix of distance between your variables, for exemple with cor and do a hierarchical clustering with hclust .
• Do a pca then cluster the projection of your variable on the created components. That way, you can use kmeans to cluster your variables.
• If all your variables are numeric, you can transpose your dataframe and do a k-mean clustering.

Also, the question of why you want to cluster the variables probably need some more thought.