An integrated classification method: Combination of LP and LDA

Abstract

Behavior analysis of credit cardholders is one of the main research topics in credit card portfolio management. Usually, the cardholder's behavior, especially bankruptcy, is measured by a score of aggregate attributes that describe cardholder's spending history. In the real-life practice, statistics and neural networks are the major players to calculate such a score system for prediction. Recently, various multiple linear programming based classification methods have been promoted for analyzing credit cardholders' behaviors. As a continuation of this research direction, this paper proposes an integrated classification method by using the fuzzy linear programming (FLP) with moving boundary and Fisher Linear Discriminant analysis(LDA). This method can improve the accurate rate in theory. In the end, a real-life credit database from a major US bank is used for explaining the idea as an example. © Springer-Verlag Berlin Heidelberg 2005.