It reduces the number of variables in a dataset by grouping them into smaller sets of correlated variables.