Study On The Outstanding Taxes Of Enterprise Based On Data Mining

With the popularity of big data technology, through use the data mining method on large number of business data to discovery knowledge which hidden in the data, these knowledge can be useful for decision support, product marketing and other applications, it can bring a lot of convenience to government work and enterprises. Using big data technology to predict the outstanding tax can protect the national tax revenue and bring a lot of convenience for the tax inspection department.This thesis began outstanding tax research based on the tax records of local taxation bureau of Chongqing, we first analyzed the characteristic of tax data such as the meaning of each field in tax records and the correlation between fields, and formulated the corresponding filtering strategy accordingly. Then we established the fact table and dimension table based on the tax records and tax data dictionary, then built tax data warehouse to take the multidimensional dimension analysis. We found that there are some relationship between enterprise tax and the industry, the area which enterprise belongs to. So we made a summary of tax data of each enterprise at each month, tax data of each industry at each month and tax data of each area at each month. We also established enterprise network based on enterprise contact such as investors or legal person. And then we took statistics of tax data of the associated enterprises at each month.In order to protect the national tax revenue, we predicted the outstanding tax possibility of one enterprise based on a period of time of tax data before now. This thesis used the classification prediction idea based on the data mining to build prediction model. We took statistics of enterprise tax data such as enterprise tax summary data at each month in the observation time window, industry tax summary data at each month in the observation time window and tax bureau tax summary data at each in the observation time window as feature attribute to predict whether outstanding taxes behavior happen in next month.In this paper, we took our performance contrast of classification prediction model with choosing different experimental data set and comparing the performance before and after attribute selection, setting different numbers of observation time windows and choosing different classification algorithm. The experimental results show that the performance of Random Forest classification algorithm based on decision tree has the best performance, accuracy, recall rate and F value can reach 90%, the outstanding tax of enterprise can be predicted with the tax data of enterprise, the tax data of industry which enterprise belongs to and the tax data of area which enterprise belongs to a period of time before, the method of feature construction put forward in this article has strong adaptability and expansibility.