Ulbricht, Jan; Tutz, Gerhard
Combining Quadratic Penalization and Variable Selection via Forward Boosting.
Department of Statistics: Technical Reports, No.99
Quadratic penalties can be used to incorporate external knowledge about the association structure among regressors. Unfortunately, they do not enforce single estimated regression coefficients to equal zero. In this paper we propose a new approach to combine quadratic penalization and variable selection within the framework of generalized linear models. The new method is called Forward Boosting and is related to componentwise boosting techniques. We demonstrate in simulation studies and a real-world data example that the new approach competes well with existing alternatives especially when the focus is on interpretable structuring of predictors.