Mixed integer programming constrained discrimination model for credit screening
Paul Brooks, Eva Lee · Spring Simulation Multiconference · 2007
Classification, the development of rules for the allocation of observations to groups, is a fundamental machine learning task. A classic example is an automated system for a lending institution that decides whether to accept or reject a credit application. One might desire a machine that allows the non-classification of certain observations that exhibit attributes of belonging to more than one group. This option would allow inspection by an expert for difficult cases, or serve as an indication that more data needs to be collected. Classification with an option to reserve judgment on an observation is known as constrained discrimination.We consider a two-stage model for multi-category constrained discrimination in which limits on misclassification rates of training observations may be pre-specified. The mechanism by which the misclassification limits are satisfied is a rejection option, also known as a reserved judgment group, for observations not demonstrating properties of membership to any of the groups.