Cross Validation of Selection of Variables in Multiple Regression.
Abstract
Techniques and criterion for selection of the 'best' subset of variables to be used in a regression model are reviewed. A model was developed using the Automatic Interaction Detection (AID) algorithm as a pre-screening device for locating those variables most important to the regression including interaction terms. Five previous models including the one developed by AID and one developed by Westinghouse on avionic characteristic data are used in cross validation experiments to determine the predictive power of these models on a new set of data points using the same set of variables. A cross validation R(2) value is discussed as a criterion for choosing between competing models. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Dec 01, 1979
- Accession Number
- ADA080407
Entities
People
- Joseph Richard Cafarella Jr
Organizations
- Air Force Institute of Technology