| Summary: | 碩士 === 國立暨南國際大學 === 資訊管理學系 === 99 === Over the past decades, the corporate credit rating status has been extensively studied by researchers in turbulent economic environment; the ratings performances have the potential impact on bank decision-making. Traditional corporate credit rating models are employed statistical methods to estimate rating status, but the model established by statistical methods in dealing with increasingly complex data is not perform a satisfactory job. Nowadays, some researchers began to use machine learning techniques to cope with the related problem, and the machine learning approach would not satisfy strict statistical limitation. In this paper, we applied Relevance Vector Machine (RVM) and Directed Acyclic Graph (DAG) methods to deal with multi-class classification (namely DAGRVM) and the subsequent experimental results could give a reference for banker to make suitable financial granting. To overcome the opaque nature of RVM, the investigation utilized Rough Set Theory (RST) to derive intuitive decision rule from RVM. The comprehensive decision rule would enhance the practical application. Therefore, the experimental results show that the DAGRVM method is an effective technique for the classification of credit rating, and it can obtain better classification accuracy (88%) than the Directed Acyclic Graph Support Vector Machine (DAGSVM). Moreover, the rules extracted by RVs model can be effectively used as a reference for enterprises.
|
|---|
