ABSTRACT
Linear (or generalized) linear models are probably the most widely used statistical analysis tools in all areas of applications. Combined with regularization approaches (see Section 17.2) they can easily handle situations where the number of predictors greatly exceeds the number of observations (high-dimensional data). Despite their structural simplicity, (generalized) linear models often perform very well in such situations. In addition, they can be used as screening tools: the selected predictors form the basis for more complex models (e.g., models that include interactions or nonlinear functions).
