PCA and k-Means Lab

The first principal component: the direction of most variance, and the line closest to the data

direction of the line (degrees)

USArrests: 50 states and 4 variables in two principal components

biplot: states (scores) and variables (loadings)

proportion of variance explained

k-means, one step at a time (150 points from 3 groups)

number of clusters K

Where k-means struggles (best of 20 random starts, K = 2)

standardise the features first

the true groups

k-means