A classification model for a two-class (New product purchase) discrimination process using multiple-criteria linear programming

The purpose of this paper is to outline a new research stream that will focus on the development of a discriminant classification procedure for categorization of product success or failures. The success/failure of each product is based upon their concept and panel score. A key concept in linear programming-based data mining is that the misclassification can be reduced by suing two objectives. One objective is to maximize the minimum distance (MMD) of data records from a critical value. The second objective separates the data records by minimizing the sum of the deviations from the critical value (MSD). Instead of maximizing the MMD or minimizing MSD, a tradeoff between MMD and MSD is found to find a compromise solution. INTRODUCTION Discriminant analysis differs from most statistical techniques because the dependent variable is discrete rather than continuous. One might assume that this type of problem could be handled by least squares regression by employing independent variables to predict the value of the value of a discrete dependent variable coded to indicate the group membership of each observation. While this approach may work involving two groups, it can be extended to more than two groups. Methods of Estimating a Classification Model Linear Discrimination by the Mahalanobis Method The objective of discriminant analysis is to use the information from the independent variables to achieve the clearest possible separation or discrimination between or among groups. In this respect, the two group discriminant analysis is no different from multiple regression. We use the independent variables to account for as much of the variation as possible in the dependent variable.