Research On Tax Prediction Model Analysis Based On Machine Learning

As a significant component of national finance,a more accurate tax forecast can provide a reference basis for the fiscal budgeting process and tax planning.Most current tax forecasting methods generally use machine learning models for annual tax forecasts for specific regions and require retraining models for new region forecasts,reducing tax administration’s efficiency.This paper analyzes and studies the tax forecasting models based on machine learning models from monthly tax forecasting.The main work is as follows:Firstly,use a fusion of pearson correlation coefficient(PCC)analysis and machine learning random forest(RF)feature importance(RFPCC)to jointly analyze 12 economic indicators,including tax revenue,gross domestic product,import and export trade,freight volume,and per capita disposable income.The analysis screens the indicators to determine the economic indicators that have a greater impact on the tax revenue,realizes the screening and dimensionality reduction of the original data,and uses the determined economic indicators as the input to the model to forecast the tax revenue.By simulating the particle swarm optimization(PSO)algorithm and the whale optimization algorithm(WOA),the convergence accuracy of the PSO algorithm is better than that of the traditional PSO algorithm.The least squares support vector regression(LSSVR)tax forecasting model based on RFPCC and WOA is constructed to improve the forecasting performance of the machine learning model by using WOA to optimize the parameters of the LSSVR model.To improve the prediction performance of the machine learning model,the parameters of the LSSVR model are optimized using WOA.The optimized model is compared with other tax forecasting models,and the experimental results show that the given RFPCC-WOA-LSSVR model has better tax forecasting results for the Guangxi region,such as the average absolute percentage error value is 8.78%,which verifies the effectiveness and practicality of the model.Secondly,another PSO-BP neural network machine learning tax prediction model is presented based on the fusion of the chaos(C)model and simulated annealing(SA)algorithm.The C model is used to optimize the local optimal solution of the PSO algorithm and the SA algorithm to optimize the global optimal solution of the PSO algorithm,and the C model combined with the SA algorithm is constructed to improve the PSO algorithm.The experimental results show that the CSAPSO-BP neural network prediction model can be better fitted to the tax data of the Guangxi district and the tax data of Hubei province,such as the average absolute percentage error values are 7.28% and 9.33% respectively,and the evaluation indexes are the lowest.Finally,a machine learning tax forecasting model system is designed for easy and efficient tax forecasting.The system is based on MATLAB’s graphics user interface(GUI)and contains the RFPCC-WOA-LSSVR model and CSAPSO-BP neural network model to forecast the input data of the above model system.The system is easy to operate and easy to maintain.