Stable Rules Evaluation for a Rough-Set-Based Bipolar Model: A Preliminary Study for Credit Loan Evaluation

摘要

The modern business environment is full of uncertain and imprecise circumstances that require decision makers (DMs) to conduct informed and circumspect decisions. In this regard, rough set theory (RST) has been widely acknowledged as capable to resolve these complicated problems while relevant knowledge can be extracted—in the form of rules—for decision aids. By using those learned rules, an innovative bipolar decision model that comprises the positive (preferred) and negative (unwanted) rules, can be applied to rank alternatives based on their similarity to the positive and the dissimilarity to the negative ones. However, in some business cases (e.g., personal credit loan), applicants need to provide information (values) on all the attributes, requested by a bank. Sometimes, experienced evaluators (e.g., senior bank staff) might question the validity of some values (direct or indirect evidences) provided by an applicant. In such a case, evaluators may assign additional values to those attributes (regarded as non-deterministic ones) in a bipolar model, to examine the stability of a rule that is supported by questionable instances. How to select those rules with satisfactory stability would be an important issue to enhance the effectiveness of a bipolar decision model. As a result, the present study adopts the idea of stability factor, proposed by Sakai et al. [1], to enhance the effectiveness of a bipolar decision model, and a case of credit loan evaluation, with partially assumed values on several non-deterministic attributes, is illustrated with the discussions of potential application in practice.