Post Mining of Multiple Criteria Linear Programming Classification Model for Actionable Knowledge in Credit Card Churning Management

In this paper, we study how to switch the customers from an undesirable class to a desirable one in credit card churning management by post mining. Multiple Criteria Linear Programming (MCLP) classification model, an optimization-based data mining method, is firstly used to classify the samples. In post mining phase, we build a case base formed by a series of typical positive instances for the entire negative population as their "good examples". These positive instances are on or near the boundary between the two classes, and thus closest to negative objects to ensure lowest switching cost. Switching plan for each negative object is then generated based on the case base, according to minimum cost principle. Real dataset from a large commercial bank of China is used to validate the method we proposed.