An Information Processing Method Oriented To Sampling Of Tax-Checking

Sampling of Tax-Checking is a process that the tax authorities find at uttermost problems and doubts existed in enterprises'tax report and return process, the suspected tax-evasion enterprises, and identify tax-checking object by means of analyzing the general financial data information and the tax return data information regularly reported to the tax agency by enterprises. The core content of the research on information processing of sampling of tax-checking is making pattern recognition on the state of enterprises'tax return. The research result will directly make impact on making a reasonable plan of tax-checking. Therefore, sampling of tax-checking is an important research field of Tax.There are many problems exist in the sampling of Tax-Checking. Such as: there are large-scale tax-related financial data of enterprises, the sampling method largely depend on the reliability of samples, and it is difficult to reflect the strong non-linear relationship between enterprises'tax return and tax evasion. To address these problems, in this article we introduce the Support Vector Machine (SVM) in machine learning and neural network clustering method into information processing of sampling of tax-checking, which can identify the tax-checking object list through classifying judgment and clustering analysis on tax-related data samples of enterprises.On the basis of analyzing and researching the theory and the implementation method on Support Vector Machine, Genetic Algorithm (GA) and Clustering Analysis, the article makes use of the idea of system engineering to design the information processing model of sampling of tax-checking which is based on the combination of Support Vector Machine and Self-organizing feature map (SOM). First of all, judge and classify whether taxpayers are included in checking scope by making use of good classifying ability of Support Vector Machine, according to the historical information of collection and management on taxpayers. While implement the specific classification, these factors that the kernel parameters and the regularization parameter C of Support Vector Machine is difficult to define must be fully considered. Furthermore, Genetic Algorithm is introduced, making use of the good global search ability of Genetic Algorithm to define the parameters of Support Vector Machine automatically. Then, based on the characteristics that taxpayers with similar tax evasion means must have similar sample properties, taxpayers which are included in the scope of sampling of tax-checking can be clustered through the method of combining Self-organizing feature map and k-means clustering algorithm. It provides the reference for making a reasonable tax-checking plan, as well as the inspection focus of tax-checking implementation, which is more beneficial to tax-checking implementation.In the clustering method which combines Self-organizing feature map and k-means algorithm, we at first make use of SOM algorithm for rough clustering, then initialize the initial cluster center of K-means algorithm by the link weight of SOM, and at last cluster samples precisely by K-means algorithm. The experiment results show that this kind of combination of the clustering algorithm can enhance the accuracy of clustering.In the end, combined with the index of sampling of value-added tax- checking, through instance data, the article validate the effectiveness of the information processing model of sampling of tax-checking based on GA-SVM and SOM-K.