Prediction Of Non-VAT Owed Tax Based On Deep Learning In Large Enterprises

With the continuous development and application of artificial intelligence,through in-depth learning and other methods to analyze and predict a large number of data,the results will be applied to all areas of need,for each of the required fields to bring a lot of convenience.This paper makes use of the depth learning algorithm to forecast the non value-added tax in large enterprises based on the tax data,which can better serve the tax inspectors,improve the quality and efficiency of the inspectors ' decision,and guarantee the tax revenue of the country.When analyzing tax data,the data volume is huge and complex,it is difficult to analyze the data directly,so it is necessary to deal with the raw data in the tax analysis and forecast,establish the data warehouse conforming to the business logic and satisfy the demand,which is the basis of the tax information data and basic information data of large enterprises in the South Sea area Combined with professional tax knowledge to preprocess data,clean out illegal data,redundant data,error data,and other forms of noise data,establish a non-VAT tax Data warehouse which accords with the fact table and multiple dimension table of this article.Then,based on the data in the non-VAT tax Data Warehouse,we consider the characteristic attributes that affect the enterprise's tax due to many dimensions.It is found that the tax-owed behavior of enterprises has a certain time correlation with geographical area,industry type and registration type,and finally,combined with relevant knowledge and experience provided by tax personnel,it can determine multiple attribute attributes affecting enterprise's tax.It contains financial data,taxpayer information,tax data.Finally,according to the characteristic attribute data that affect the enterprise's owed tax in the previous year,the paper forecasts whether the tax-owed behavior will occur in the next year,and studies the model of the enterprise's owed tax based on the depth learning algorithm,and designs the model input layer,the hidden layer and the output layer.Using the 2015 and 2016 non-VAT tax related data provided by the tax authorities of Nanhai district of Foshan,the accuracy of the forecast model is validated,and the accuracy rate of each step optimization is recorded and analyzed respectively,and the highest average prediction accuracy rate is now 88.45%,Compared with traditional methods such as artificial prediction and measurement,the accuracy rate of less than 80% is greatly improved,and the use of multidimensional data is more persuasive,which proves that the model based on depth learning algorithm is reasonable and reliable,and provides scientific and rational research ideas for tax authorities.