纳税评估的广义回归神经网络建模与实证

发布时间：2018-03-10 06:31

  本文选题：纳税评估　切入点：广义回归神经网络　出处：《系统工程》2015年11期 　论文类型：期刊论文

【摘要】：针对上海市某区386家中小企业15个财务指标数据,运用灵敏度分析方法筛选出对判定纳税情况具有显著影响的10个评价指标,采用自组织神经网络方法把全部386个样本分成性质相似的训练样本、检验样本和测试样本,通过逐步减小光滑因子值确定其合理值,建立纳税评估广义回归神经网络(GRNN)模型。与线性回归、判别分析、Logistic和支持向量机等模型的结果对比表明:GRNN模型的分类错误率最低,检验样本和测试样本的II类和I类分类错误率分别低于5.4%和2.0%,平均分类错误率低于2.5%.对另外339家企业纳税情况的判定结果表明,建立的GRNN模型具有很好的泛化能力和鲁棒性。
[Abstract]:According to the 15 financial index data of 386 small and medium-sized enterprises in a certain district of Shanghai, the sensitivity analysis method is used to screen out 10 evaluation indexes that have a significant impact on the determination of tax payment. Using self-organizing neural network method, all 386 samples are divided into similar training samples, test samples and test samples, and the reasonable value is determined by gradually reducing the smooth factor value. A generalized regression neural network (GRNN) model for tax assessment was established. The comparison with linear regression, discriminant analysis logistic and support vector machine shows that the classification error rate of the two models is the lowest. The class II and class I classification error rates of test samples and test samples are lower than 5.4% and 2.0, respectively, and the average classification error rate is lower than 2.5.The results of tax assessment for another 339 enterprises show that the GRNN model has good generalization ability and robustness.
【作者单位】： 上海商学院财经学院;上海理工大学管理学院;
【基金】：上海高校知识服务平台“上海商贸服务业知识服务中心”建设子项目“税收风险管理信息系统设计及开发”(ZF1226) 上海高校重点学科“商务经济学”建设项目,参加本课题的还有尹淑平、张娇芳、史昱民、包时军、邬春学和高丽萍等同志
【分类号】：F812.42
，

本文编号：1592162