基于BP神经网络的个人信用风险评估模型的研究

发布时间：2018-01-07 15:08

本文关键词：基于BP神经网络的个人信用风险评估模型的研究　出处：《太原理工大学》2017年硕士论文　论文类型：学位论文

【摘要】：随着经济社会的发展进步,我国也逐渐向“信用经济”方向迈进。加之我国经济政策的优化调整,信贷业务已逐渐在我国的金融市场上遍地开花。随着城市化建设步伐的加快,更加推动了“房贷”、“车贷”等个人信贷业务的井喷式发展。但毕竟我国的信用体系建设起步晚、基础差、经验少,在实际应用中难免有不足之处,尤其在“信用风险”控制方面常常顾此失彼。通过深入研究调研发现,目前我国大多数银行在进行个人信用风险评估时,普遍采用打分制的评估方法,且业务开办实施多年来,评估指标却未能与时俱进,基本上没有较大的改动。显然这种评估方式带有很强的主观性,评估指标也相对落后,略显刻板、单一。对此,本文提出了新的评估指标体系,并采用改进的BP算法建立了个人信用风险评估模型,希望能对银行的个人信贷业务提供积极的参考意义。基于BP神经网络的个人信用风险评估模型的研究,本质上需要解决三个问题:BP神经网络算法的改进、个人信用风险评估指标体系的建立及评估模型的构建:(1)BP神经网络算法的改进:BP神经网络受初始权值和阈值的约束,容易陷入局部极小值点。而混合蛙跳算法是一种仿生智能优化算法,具有良好的全局搜索能力,所以本文提出了用改进的混合蛙跳算法优化BP神经网络的解决办法。(2)个人信用风险评估指标体系的建立:在符合我国信用体系的实际发展背景下,本文主要参考借鉴了建设银行个人信用评级、蚂蚁金融芝麻信用、人人贷、美国FICO这四种个人信用评估方法的评估指标,以及中国人民银行征信中心出具的个人征信报告。然后在传统评估指标的基础上,剔除了一些对个人信用识别能力较差的指标项,添加了2项对个人信用评估有较强识别能力的“互联网线上”指标项,打破了以往传统评估指标仅仅局限于“线下”的桎梏,做到了对借款人实施“线上+线下”的全方位评估。(3)评估模型的构建:利用改进后的混合蛙跳算法来优化BP神经网络进行模型建立,对“线上+线下”的18项指标进行科学的预测分析,实现对个人信用风险等级的评估,从而规避了传统评估方法的主观性,缩短了评估流程,提升了评估效率。最后通过实验论证模型的预测准确率。
[Abstract]:With the development and progress of economy and society, our country is also gradually moving towards the direction of "credit economy", in addition to the optimization and adjustment of our country's economic policy. Credit business has gradually blossomed in China's financial market. With the acceleration of the pace of urbanization, the promotion of "housing loans". "car loan" and other personal credit business blowout development. But after all, the construction of credit system in China started late, the foundation is poor, the experience is less, it is inevitable to have deficiencies in the practical application. Especially in the control of "credit risk", we often lose sight of each other. Through in-depth research, it is found that most banks in our country generally use the evaluation method of scoring system when carrying out personal credit risk assessment. And the business implementation for many years, the evaluation index has failed to keep pace with the times, basically no major changes. Obviously, this evaluation method has a strong subjectivity, evaluation indicators are relatively backward, slightly rigid. Single. In this paper, a new evaluation index system is proposed, and an improved BP algorithm is used to establish a personal credit risk assessment model. The research of personal credit risk assessment model based on BP neural network, in essence, need to solve three problems: the improvement of BP neural network algorithm. The Establishment of personal Credit risk Evaluation Index system and the Establishment of Evaluation Model the improvement of the BP neural network algorithm is restricted by the initial weight and threshold. The hybrid leapfrog algorithm is a bionic intelligent optimization algorithm with good global search ability. Therefore, this paper proposes an improved hybrid leapfrog algorithm to optimize the BP neural network solution. 2) the establishment of personal credit risk evaluation index system: in line with the actual development of our country's credit system background. This paper mainly refers to the construction bank personal credit rating, ant finance Sesame credit, peer-to-peer lending, the United States FICO these four personal credit evaluation methods evaluation indicators. And the personal credit report issued by the credit information center of the people's Bank of China. Then on the basis of the traditional evaluation index, some index items with poor personal credit recognition ability are eliminated. This paper adds two "Internet online" indicators which have a strong ability to identify personal credit evaluation, which breaks the shackles of traditional evaluation indicators that are limited to "offline" only. The comprehensive evaluation model of "line up and below line" is constructed: the improved hybrid leapfrog algorithm is used to optimize the BP neural network to build the model. This paper carries on the scientific forecast and analysis to the "on-line and offline" 18 indexes, realizes the evaluation of personal credit risk grade, thus avoids the subjectivity of the traditional evaluation method and shortens the evaluation process. The evaluation efficiency is improved. Finally, the prediction accuracy of the model is proved by experiments.
【学位授予单位】：太原理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：F832.4;TP183

【参考文献】