摘要
信用如今已经渗透至社会生活、工作之中,信用评估是金融、通讯等服务行业对消费者个体的重要需求。在分析个人信用影响因素及其相关数据建模基础上,改进了应用Logistic回归建模过程中所用到的最速下降法,有效减少了回归建模过程中的迭代次数与迭代时间。原始最速下降法相邻方向是正交的,导致越是靠近极值点步长越小,收敛速度慢;而改进后的最速下降法通过结合上一次的搜索方向确定当前搜索方向,改变了原本锯齿形的曲折搜索路径。为验证所提出方法的有效性和可行性,围绕迭代次数与迭代时间进行了实验验证。验证实验结果表明,改进的最速下降法减少了计算过程中的迭代次数,从而提高了运算效率;针对影响信用数据提供不全的记录,将转移概率矩阵应用于信用评估,可解决未来信用预测评估问题。
Credit has been combined closely with people＇ s daily life and work. And credit assessment maintains a significant requirement of customers in service industries such as finances and communications. In this paper, the Steepest Descent Method （SDM） in Logistic Regression analysis has been improved based on influence factors of credit and relative data of modeling ,reducing iteration times and time in regression modeling. The strategy can be explained that in original SDM, adjacent searching directions keep orthogonai and steps ap- proach zero when they are close to the extreme point, which contributes to a slow rate of convergence. Yet,in the improved scheme,cur- rent searching direction has been determined by the last one and zigzag directions are eliminated therefore. In the experiments, it is proved that times of iterations is decreased and computational efficiency is enhanced. Moreover, aiming at defective credit records, matrix of tran- sition probability has been adopted in order to solve problem of the credit assessment and prediction in the future.
出处
《计算机技术与发展》
2017年第8期47-51,共5页
Computer Technology and Development
基金
国家"863"高技术发展计划项目(2006AA01Z201)
关键词
信用评估
最速下降法
LOGISTIC回归
转移概率
credit evaluation
steepest descent method
Logistic Regression
transition probability
