Error Correction Method in Classification by Using Multiple-Criteria and Multiple-Constraint Levels Linear Programming

Abstract: In classification based on multiple-criteria linear programming (MCLP), we need to find the optimal solution of the MCLP problem as a classifier. According to dual theory, multiple criteria can be switched to multiple constraint levels, and vice versa. A MCLP problem can be logically extended into a multiple-criteria and multiple-constraint levels linear programming (MC2LP) problem. In many applications, such as credit card account classification, how to handle two types of error is a key issue. The errors can be caused by a fixed cutoff between a "Good" group and a "Bad" group. Two types of error can be systematically corrected by using the structure of MC2LP, which allows two alterable cutoffs. In order to do so, a penalty (or cost) is imposed to find the potential solution for all possible trade-offs in solving MC2LP problem. Some correction strategies can be investigated by the solution procedure. Furthermore, a framework of decision supporting system can be illustrated for various real-life applications of the proposed method.