Alternative discriminant vectors in LP models and a regularization method

Since the mathematical programming approach was proposed to solve discriminant analysis problems, a number of anomalies with this technique have been identified and remedied. However, the existence of alternative or multiple discriminant vectors and their impact has not been fully addressed. This article examines the significance of alternative discriminant vectors in LP models, characterizes the conditions under which they are generated, and proposes a regularization method to eliminate this undesirable outcome.