Support Vector Machine Combination Model Based On Grey Relation Analysis

Fiscal revenue is an important economic lever for the state to exercise macroeconomic control.The growth of fiscal revenue is related to a country 's economic development and social progress.Accurately predicting the growth of fiscal revenue is particularly important for the macro-control of the state and government.By understanding the development status of research on fiscal revenue forecasting at home and abroad,the necessity of demonstrating fiscal revenue forecasting is obtained.In order to determine the main indicators that affect fiscal revenue and the necessity of selecting indicators,it is proposed to use gray correlation analysis to select the main controlling factors that affect fiscal revenue.Because gray correlation analysis does not have too many requirements on the number of sample data and internal connections,this is an important reason why we select it for index dimensionality reduction.Support vector machine is the most widely used theory in statistical learning theory.It can solve practical problems such as small samples,nonlinearity and high dimensions.Introduced the status and development of the support vector machine model,as well as the principle,advantages and disadvantages of the model.The fiscal revenue of 1952-2018 and its six impact indicators were selected as sample data.Among them,1952-2014 was selected as the training sample,2014-2018 as a test sample.In addition to autoregression,another six impact indicators are added.By changing the input sample,that is,adding the fiscal revenue of the previous year on the basis of the six impact indicators,the support vector machine model is used to predict the fiscal revenue to find the best The prediction method of the support vector machine model is that the input sample is the fiscal revenue of the previous year and the support vector machine model has the smallest prediction error when the six impact indicators are predicted.And compared with BP neural network,RNN neural network and random forest model prediction error comparison analysis,it is concluded that this support vector machine model has better prediction accuracy.In order to solve the defect of incomplete information extraction by a single model and improve the accuracy of the prediction model,this paper proposes the idea of combined model prediction based on support vector machines.Given the development prospects andadvantages of the combined model,the support vector machine model is combined with the BP neural network,RNN neural network and random forest models to obtain three combined models.The key of the combined model lies in the calculation of weights,so it is used The two weight calculation methods,the arithmetic average method in constant weights and the optimal weight method proposed by Yang Tingfang and others in change weights,use three combined models for prediction respectively,and combine their combined models and support vector machine models.Perform comparative analysis.After many model experiments,the combined model based on support vector machine and BP neural network not only has better prediction effect than single model,but also has smaller error than other combined models.The advantage of support vector machine is that it can solve the problems of small samples,nonlinearity and high dimensionality,and can guarantee the global optimal solution.Neural networks will have problems such as local maximum,over-learning and dimensionality disasters.These are also two types.The reason why the model can be well integrated.Using this combination method,the maximum advantages of the two models are exerted,and it is concluded that this combination model can achieve better accuracy and reliability than the single model for fiscal revenue prediction.