Research On The Identification Of Added-value Tax Behavior In Small And Medium-sized Enterprises Based On Machine Learning Algorithm

With the rapid development of China's market economy,small and medium-sized enterprises have also been growing in the environment of economic development.How to carry on the reasonable tax collection and management to these small and medium-sized enterprises has become the focus of the tax department.With the rapid development of big data,artificial intelligence and network information technology,the traditional tax inspection methods need innovation and improvement.The integration of enterprise tax and big data machine learning algorithm can not only discover more features behind the data,but also improve the work efficiency while bringing convenience to the actual tax collection.Therefore,how to combine the financial information of small and medium-sized enterprises' tax payment with the data related to their widespread multi-source tax payment behavior,and how to use big data machine learning algorithm to carry out intelligent screening research on small and medium-sized enterprises' VAT tax payment behavior is of great practical significance to strengthen the tax management monitoring function and realize the intelligent tax inspection of small and medium-sized enterprises.Based on the review of the relevant concepts,theories,methods and literature of the VAT behavior screening of small and medium-sized enterprises firstly,this paper carries out the research according to the data scientific process of "data acquisition ? data cleaning ? data exploration ? learning method modeling ? result discussion and display" of the VAT behavior screening of small and medium-sized enterprises.The main work and research conclusions are as follows:(1)Data preparation and feature selection.First of all,taking the value-added tax behavior of small and medium-sized enterprises in H city of Shandong Province as the research object,244 samples of annual financial statements(enterprise balance sheet,profit statement,cash flow statement)and tax return data of 85 enterprises in 2016-2018 were collected and sorted,and exploratory analysis was made on the value-added tax behavior of small and medium-sized enterprises;Secondly,the sample data was randomly divided into 70%: 30% It is divided into training set and test set to implement machine learningalgorithm modeling and validation research;Finally,based on the five concept indicators of corporate profitability,corporate solvency,corporate operation ability,corporate growth ability and corporate cash ability,29 financial indicators,including sales profit rate,sales gross profit rate and sales net interest rate,are selected to build a small and medium-sized enterprise value-added tax paying bank For the screening index system.The dependent variable is the small and medium-sized enterprises' value-added tax payment behavior.Use "0" to indicate the enterprises' tax compliance,use "1" to indicate the enterprises' tax noncompliance.There are 45 samples of tax noncompliance in the data.(2)Machine learning algorithm exploration and model integration verification.The first mock exam is the first mock exam.First,the missing values and the standardized processing of sample data are processed.Secondly,the logistic regression,decision tree and support vector machine are used to classify the sample training set data by single model training test.Based on the accuracy,precision,recall and F-measure,the learning results of each single model are comprehensively evaluated on the test set;finally,the three learning strategies are applied to the sample set.Three classification algorithms are used as the basic model,and stacking integration method is used for integrated learning training and test verification.The expected effect of small and medium-sized enterprises' VAT tax behavior screening integration model is obtained.It is found that the single screening model based on logistic regression,decision tree and support vector machine can achieve certain effect,and the performance of decision tree and support vector machine is better among the three classification learning algorithms;however,from the comprehensive analysis of accuracy index and AUC results,it is found that the integration of multiple weak classification based models can significantly increase the performance of the model.Therefore,the integrated model based on machine classification learning algorithm has a wide application prospect in the intelligent learning of small and medium-sized enterprises' VAT behavior screening.