Feature transformation is a group of methods that create new features (predictor variables). The methods are useful for dimension reduction when the transformed features have a descriptive power that is more easily ordered than the original features. In this case, less descriptive features can be dropped from consideration when building models.
Feature transformation methods are contrasted with the methods presented in Feature Selection, where dimension reduction is achieved by computing an optimal subset of predictive features measured in the original data.
The methods presented in this section share some common methodology. Their goals, however, are essentially different:
Nonnegative matrix factorization is used when model terms must represent nonnegative quantities, such as physical quantities.
Principal component analysis is used to summarize data in fewer dimensions, for example, to visualize it.
Factor analysis is used to build explanatory models of data correlations.