The OSCAR for Generalized Linear Models

Petry, Sebastian und Tutz, Gerhard (2011): The OSCAR for Generalized Linear Models. Department of Statistics: Technical Reports, No.112

Abstract

The Octagonal Selection and Clustering Algorithm in Regression (OSCAR) proposed by Bondell and Reich (2008) has the attractive feature that highly correlated predictors can obtain exactly the same coecient yielding clustering of predictors. Estimation methods are available for linear regression models. It is shown how the OSCAR penalty can be used within the framework of generalized linear models. An algorithm that solves the corresponding maximization problem is given. The estimation method is investigated in a simulation study and the usefulness is demonstrated by an example from water engineering.

• Item Type: Paper
• Keywords: Variable Selection, Clustering, OSCAR, LASSO, Generalized Linear Models
• Faculties: Mathematics, Computer Science and Statistics > Statistics > Technical Reports
• Subjects: 500 Science > 500 Science
• Language: English