EVALUATION AND MANAGEMENT OF CREDIT RISK USING COMPUTATIONAL INTELLIGENCE AND MULTI-OBJECTIVE DECISION MAKING

在投资、企业融资等金融活动中，对企业经营失败的适度信用风险进行评估是非常重要的。本研究的目的是构建一个利用计算智能评估企业信用风险的系统，并通过多准则决策，在一定的允许风险下对风险进行管理，以获得最大的利润。首先，在定性和定量数据集的基础上，应用支持向量机（svm）对企业信用风险进行评估；这些业务失败的数据集有一些不平衡：失败数据只有很少的百分比，几乎所有的数据都是非失败的。为了克服这一问题，利用多目标规划和目标规划对支持向量机进行了改进。结果表明，改进后的支持向量机对元素极少的类别具有较好的分类能力。此外，利用粗糙集理论从得到的支持向量中提取简单规则和更显式的规则。其次，开发了适应环境变化的动态学习机器。学习机器可以通过增量学习来提高自己的能力。然而，如果只进行增量学习，则决策规则会变得越来越复杂，导致泛化效果较差。因此，在目前的情况下，需要去除不必要的（或障碍）数据。这叫做“遗忘”。在这项研究中，开发了几种遗忘方法，不仅以被动的方式（数据的影响随着时间的推移而减少），而且以主动的方式（发现并主动消除不必要的（障碍）数据）。如果我们只是规避风险，我们就不能进行积极的金融活动，因为每一项金融活动都有一定的风险。最后利用数据包络分析（DBA）来评价决策单元的效率，以便在一定的允许风险下获得尽可能多的利润。由于传统的DEA是基于数据集的凸包并考虑了线性值判断，因此不能适用于非线性值判断的问题。提出了一种广义DEA方法，试图对几种非线性价值判断下决策单元的效率进行测度。通过实例证明了广义DEA的有效性。少

In financial activities such as investment and business finance, it is important to evaluate moderately credit risk against business failure of enterprises. The aim of this research is to construct a system for evaluating credit risk of enterprises using computational intelligence and for managing the risk to get as much profit as possible under some allowable risk by virtue,of multiple criteria decision making.Firstly, support vector machines(SVMs) were applied to evaluate credit risk of enterprises on the basis of qualitative and quantitative data sets. Those data sets for business failure have some unbalance : failure data are only a few percentage, and almost of all date are of nonfailure. In order to overcome this problem, SVMs were modified by using multi-objective programming and/or goal programming. As a result, the modified SVM showed a good classification ability for the category with extremely fewer elements. Moreover, the rough set theory was applied to extract simple and e … More xplicit rules from the obtained support vectors.Secondly, dynamically adapting learning machines for the change of environment were developed. Learning machines can increase their ability by making incremental learning. If we make only incremental learning, however, the decision rule becomes more and more complex, which resluts in poor generalization. Therefore, it is needed to remove unnecessary(or obstacle) data under the present situation. This is called "forgetting". In this research were developed several methods for forgetting not only in a passive manner in which the influence of data decreases over time but also in an active way in which unnecessary(obstacle) data are found and removed actively.If we only avoid risk, we can not make an active financial activities, because every financial activitiy has some risk. Finally, therefore, data envelopment analysis(DBA) was applied to evaluate the efficiency of decision unit in order to get as much profit as possible under some allowable risk. Since the conventional DEA is based on the convex hull of data set taking into account the linear value judgment, it can not be applied to problems under nonlinear value judgment. A generalized DEA was developed to attempt to measure the efficiency of decision unit under several kinds of nonlinear value judgments. The effectiveness of the generalized DEA was proved through several examples. Less