Knowledge extraction from multiple criteria linear programming classification approach
Yuejin Zhang, Peng Zhang, Lingling Zhang, Yong Shi · Procedia Computer Science · 2010
As an effective model for classification, Multiple Criteria Linear Programming (MCLP) has been widely used in business intelligence. However, a possible limitation of MCLP is that it generates unexplainable black-box models which can only tell us results without reasons. To overcome this shortage, in this paper, we present a knowledge mining strategy which mines explainable decision rules from black-box MCLP models. Firstly, we use the rough set theory to distinguish the definable set where samples are perfectly classified, from the rough set where misclassified samples may exist. Then, to get explainable knowledge, we present a clustering-based decision rule extraction approach to extract knowledge from the definable set, and a rough set-based rule extraction approach to the rough set. Finally, empirical studies on real world VIP Email data sets demonstrate that our method can effectively extract explicit rules from MCLP model with only a little lost in performance.