Multivariate outliers

The preceding example has just given an example of looking at outliers from a univariate, or single variable, perspective. However, outliers can also occur in a multivariate or combination form. In these cases, visualizing outliers in two dimensions can be a start, but as the dimensionality increases they can be more difficult to isolate. For multivariate outliers you can use distance and influence measures such as Cook's D or Mahalanobis distances to measure how far they are from a regression line. Principal component analysis can also help by reducing the dimensionality first, and then examining the higher order principal components that could include the outliers.

Get Practical Predictive Analytics now with O’Reilly online learning.

O’Reilly members experience live online training, plus books, videos, and digital content from 200+ publishers.