Variable Selection Deviation Measures
2015-03
Loading...
View/Download File
Persistent link to this item
Statistics
View StatisticsJournal Title
Journal ISSN
Volume Title
Title
Variable Selection Deviation Measures
Authors
Published Date
2015-03
Publisher
Type
Thesis or Dissertation
Abstract
Many exciting results have been obtained on model selection methods for high- dimensional data in both efficient algorithms and theoretical developments. The powerful penalization methods for the variable selection in both regression and classification can give sparse representations of the data even when the number of predictors is much larger than the sample size. One important question then is: How do we know when a sparse pattern identified by a model method is reliable? In this dissertation, we propose variable selection deviation measures that give one a proper sense on how many predictors in the selected set are likely trustworthy in certain aspects. In the first part of this thesis, we investigate the instability of the penalization methods (Lasso, SCAD, MCP and Stability Selection) in term of the variable selection. Three instability measures are studied: sequential instability, parametric bootstrap instability and perturbation instability. Then, we propose a variable selection deviation measure (VSD) to quantify the uncertainty of the selected sparse set. Simulation and a real data example demonstrate the utility of the VSD measures for application. In the second part, we propose the VSD measures for the generalized linear model (GLM), in particular, logistic regression. The VSD measures rely on good weights on the models and they help quantifying the deviation of the selected model from the true model. For the generalized linear model, we adopt the ACM (Yang (2000)) to define the weight function for GLM VSD measures. We also propose the weight function and algorithm of the VSD for Poisson regression. We implement the VSD measures on simulated dataset and four real data examples. We build an R package called glmvsd to calculate the VSD measures. After providing the target model that user wants to evaluate, this package will calculate the VSD measures according to different weight functions defined in this thesis. The package can also calculate the three instability measures for several model selection methods. In Chapter 4, the manual of this package is presented.
Keywords
Description
University of Minnesota Ph.D. dissertation. March 2015. Major: Statistics. Advisor: Ying Nan. 1 computer file (PDF); vii, 74 pages.
Related to
Replaces
License
Collections
Series/Report Number
Funding information
Isbn identifier
Doi identifier
Previously Published Citation
Other identifiers
Suggested citation
Nan, Ying. (2015). Variable Selection Deviation Measures. Retrieved from the University Digital Conservancy, https://hdl.handle.net/11299/191484.
Content distributed via the University Digital Conservancy may be subject to additional license and use restrictions applied by the depositor. By using these files, users agree to the Terms of Use. Materials in the UDC may contain content that is disturbing and/or harmful. For more information, please see our statement on harmful content in digital repositories.