Model Selection with Data-Oriented Penalty.
Abstract
We consider the model selection or variables selection in the classical regression problem. In the literature, there are two types of criteria for model selection, one based on prediction error (FPE) and another on information theoretic considerations (GIC). Each of these criteria uses a certain penalty function which is the product of the number of variables j in a submodel and a function C(n) depending on n and satisfying some conditions to guarantee consistency in model selection. One of the important problems in such a procedure is the actual choice of C(n) in a given situation. In this paper we show that a particular choice of C(n) based on observed data, which makes it random, preserves the consistency property and shows improved performance over a fixed choice of C(n).
Document Details
- Document Type
- Technical Report
- Publication Date
- Apr 01, 1997
- Accession Number
- ADA324872
Entities
People
- Calyampudi Radhakrishna Rao
- Yipeng Wu
- Z. D. Bai
Organizations
- Pennsylvania State University