Research And Application Of Data Mining Technology In Tax Administration

Data mining is a technology that it can find the relationship in model and data from massive data using analysis tools, or extract potential, unknown and useful information, patterns and trends from massive data. The purpose is to improve decision-making capacity, detect abnormal patterns, predict future trends based on past experience. So data mining technology can be used in national taxation in many ways, like using time series technology we can predict the scale of taxation's revenue, it can help taxation administration department manage the source of taxation better and establish an effective management system to reduce the loss of taxation's revenue; using cluster analysis we can classify the taxpayers, it can improve the efficiency and quality of taxation administration, encourage taxpayers to pay tax in good faith and promote the taxpayer to establish the image of taxation credit consciously; using factor analysis method we can appraise the performance of taxation organization comprehensive, and take the combination of rewards and punishments that it can help taxation managers enhance the quality of service.This article focuses on data mining technology in tax management's application. Its purpose is to use the historical data adequately and to provide new research ideas for national taxation administration from the research of association rule mining, analysis and forecast in data mining. The prime task is:1. By comparing the advantages and disadvantages of Apriori algorithm used widely and combining the characteristics of taxation data, we study the feasibility and necessity on the applications of association rule mining in the tax system; then we study the basic principles of PSO and neural network and analyze the feasibility of combination. It lays a theoretical foundation of neural particle swarm model in the prediction of value-added tax. It provides a new way about mining research and application in the taxation system.2. It uses improved algorithm of mining association rules in taxation inspector, taxation analysis and taxation law execution, the first improved algorithm G-Apriori is applied to tax inspector. The paper uses it into mining the association rules of tax offence from the historical data and receives some reasonable rules. These rules have guiding significance in inspector job. The second algorithm is interdimension association rule based on revenue system DC-Apriori is proposed. Because data-cube can model and observe the data using multi-dimensional, it is suitable for mining multi-dimensional association rules. In the tax system, we mine multi-dimensional association rules from revenue data. Using the rules, the tax system management can make decisions from different angles for taxation analysis. So the algorithm based on data-cube has some relevance and higher efficiency to some extent. The third developed algorithm that based on Hash and AVM--H-AVM algorithms. It converts transaction data base into bite vector, get itemsets support degree by the vectors"and"operation, and just one time scan in database. Within Hash technology, it build frequent 2-itemsets by scanning database one time, so it improves the efficiency of algorithm. The experiments result tells us H-AVM algorithm proved the efficiency of association rules mining. Then we apply this algorithm in taxation law execution, it found the valuable reference information for taxation system to enhance tax law execution.3. It uses RBF neural network and particle swarm optimization theory as a tool, about the inertia weight's adjustment of PSO, we select linear adjustment strategies with small step to prevent particle swarm optimization process into the local optimum value and then build the neural prediction model using particle swarm to predict the value-added tax. We analyze from the experiment, the model can make more accurate prediction, so it can provide a more reliable basis for decision-making.