On Selecting Variables in Succession for Best Discrimination

摘要

When several variates are available to classify the individual into either of two (or more) populations, the usual practice is to construct a linear discriminant function with all the variates and reject the variates with low discriminating power. Clearly this method involves extra computations and cost. A method is suggested of selecting the variates in succession for best discrimination, when the correlation between any pair of variates is the same. (Author)