An approximate solution classification model through optimization techniques

In this work we consider the problem about classification of objects (products, individuals, companies) in two or more groups. The traditional approach is based mainly on models like Logit or Discriminant Analysis. It results in this way, either a cutting score or a measure of distance from the center of each group, defined with the simultaneous use of all the sample. In this model we consider that the classification variables are limited, inferiorly or superiorly. Through this hypothesis, it results a border for each group, that is the limit of the group. We intend to determine the border for the group and, also, a value of probabilistic nature for the measure (score) of each unit, relatively to the boundary of the group that it belongs. After that, we do the classification of a new unit, in the group where it presents the greater measure.