R Statistics Cookbook

How to do it...

We will follow a similar to the one as we have used so far. We will generate a dataset, contaminate it, and then extract both regular principal components and robust ones. Because principal components are calculated using the covariance matrix, they suffer from serious sensitivity to outliers:

1. First, we generate a synthetic multivariate Gaussian dataset:

library(MASS)library(rospca)set.seed(100)matrix = diag(10)matrix[2,1] = 0.8matrix[1,2] = 0.8matrix[4,3] = 0.8matrix[3,4] = 0.8matrix[5,6] = 0.8matrix[6,5] = 0.8

2. We now fit both principal component analysis and robust principal component analysis models:

d <- mvrnorm(n = 1000, mu=rep(0,10), matrix) prcomp(d,scale=TRUE,center=TRUE) 

These eigenvalues can be interpreted as ...