A unified description of classical classification methods for multicollinear data

In this paper a unified description of classification methods in situations with multicollinear data is proposed. It is shown that a number of the well-established methods can be derived by substituting different modified versions of the covariance matrix into either the classical Bayes method or Fisher’s linear (canonical) discriminant method. A parametric version of this modified covariance matrix is proposed. Each method corresponds to a particular value of the parameters.