碩士 / 輔仁大學 / 應用統計學研究所 / 97 / To effectively control the overdue accounts, the cash card issuing banks typically spend a lot of money to minimize the overdue losses based on the personal information. These information or variables were generally provided by the customers. However, the associated maintenance cost will dramatically increase when the number of cash card customers increase. In addition, it becomes difficult to interpret the model’s predictions when the variables are too many. This study aims to develop prediction models for cash card overdue customers. The prediction models can be used to help cash card issuing banks correctly issue the cash card to the right persons. As a consequence, the cash card issuing banks are able to minimize the overdue losses based on these predictions. To build up the predictions model, this study employs the logistic regression and neural network approaches in the first stage. In the second stage, the important variables are selected by decision trees method, and then the logistic regression and neural network prediction models are constructed based on those important variables. The prediction performances among those prediction models are compared and reported. The research findings indicate that the two-stage integrated model, which was developed through the use of seven out of nineteen variables, has a good prediction capability.
Identifer | oai:union.ndltd.org:TW/097FJU00506028 |
Date | January 2009 |
Creators | Tien-Yu Chen, 陳天佑 |
Contributors | Yuehjen E. Shao, 邵曰仁 |
Source Sets | National Digital Library of Theses and Dissertations in Taiwan |
Language | zh-TW |
Detected Language | English |
Type | 學位論文 ; thesis |
Format | 67 |
Page generated in 0.0211 seconds