BP算法在信用风险分析中的应用  

Application of BP algorithm in credit risk analysis

作　　者：庞素琳[1] 黎荣舟[2] 徐建闽[2] 

机构地区：[1]暨南大学数学系,广东广州510632 [2]华南理工大学交通学院

出　　处：《控制理论与应用》2005年第1期139-143,共5页Control Theory & Applications

基　　金：广东省自然科学基金资助项目(31906);广州市科技局科技攻关项目(2004Z3-D0231);广东省科技厅科技攻关项目(2004B10101033).

摘　　要：建立了基于BP算法的神经网络信用风险评价模型,用来对我国某国有商业银行2001年80家贷款企业进行两类模式分类.按照企业的财务状况、经营状况以及过往的信用记录分为"信用好"和"信用差"两个小组.对于每一家贷款企业,主要考虑能反映该企业的还款能力、盈利能力、经营效率和资本结构等7个财务比率作为分析变量.对该BP网络分别训练100次、390次和800次.仿真结果表明,当训练800次时,网络达到一定的稳定状态,目标函数值达到最优,分类准确率达到98.75%.此外,还给出了该BP网络的学习算法和步骤.A credit-risk evaluation model is established,which is based on back-propagation (BP) algorithm.The model has been applied to evaluate the credits of 80 applicants in a commercial bank of our country in 2001.These data are separated into two groups:a 'good credit' group and a 'bad credit' group according to their finance,management and previous credit records.As to each applicant,seven financial rates are considered that can reflect its debt paying ability,profitability,quality of management and capital structure. The BP network is trained 100,390 and 800 times respectively.The simulations show that,when the network is trained 800 times,it enters steady state and the performance function reaches optimal value,and the classification accuracy rate is 98.75%.In addition,a learning algorithm and steps of the BP network are presented as well.

关 键 词：BP算法 信用风险评价模型 信用风险分析 

分 类 号：F830[经济管理—金融学]