基于广义回归神经网络的税务稽查选案实证研究

An Empirical Study on the Sampling of Tax Audit Based on General Regression Neural Network

针对企业纳税稽查选案,采用全部样本和五重-交叉检验法(CV)分别建立线性回归、判别分析、Logistic、支持向量机(SVM)和广义回归神经网络(GRNN)模型,比较研究不同模型的建模结果。GRNN模型结构简单,训练速度快,能很好地进行小样本、连续非线性系统建模。实证研究结果表明,GRNN模型非常适用于税务稽查选案研究,在上述五种模型中,分类错误率最低,小于10%。

Based on the total sample data and the five-fold cross-validation method, the paper respectively establishes the models of multivariate linear regression （ MLR）, Logistic, linear multivariate discriminant a- nalysis （MDA）, support vector machine （SVM） and general regression neural network （GRNN） for the sampling of corporate tax audit. The GRNN prediction results are compared with those obtained with the other models. It finds that the GRNN model is characterized by simple structure and fast training algorithm, and is available for the construction of a small-sample and continuous non-linear variables system with a good prediction performance. Therefore, the GRNN model is first used for the sampling of tax audit in this paper. The empirical research results show that the GRNN model generates a less than 10% prediction classification error rate, which is the lowest among the five models.