Practical Predictive Analytics

Principal components

Principal Component Analysis (PCA) is useful in cases in which you have too many variables and you wish to capture all of the variation of data into one or two variables. Principal component does this using matrix algebra to create linear combinations of all of the variables.

PCA is also used to see which variables have the most influence within each of the principal components, so often it is a way to discard certain variables when a PCA analysis indicates certain variables have little effect on the outcome.

An example using the variables of height and weight is a good one. Since height and weight are positively correlated, choosing to use only height or only weight as an independent variable to predict body mass might ...