Multicollinearity's Effect on Classification Accuracy With Real Data Structures

• In Event: 67.048 - Multiple Linear Regression SIG: The General Linear Model

Abstract

Predictor variable multicollinearity in the use of least squares as well as maximum likelihood classification strategies – discriminant analysis and logistic regression -- are considered from the perspective of classification prediction. As opposed to prior studies that consider the effect of multicollinearity on classification accuracy by varying a constant proportion eigenvalue decrement, a method for manipulating multicollinearity while maintaining a real data set’s eigenvalue structure is used. For 26 data sets examined, it is shown that multicollinearity has no effect in respect to classification accuracy for either mathematical model.

Authors

• John D. Morris, Florida Atlantic University

• Mary G. Lieberman, Florida Atlantic University