基于支持向量机的税收预测模型研究

发布时间：2018-06-20 01:51

本文选题：支持向量机 + 税收预测　；参考：《广东工业大学》2017年硕士论文

【摘要】：针对税收预测存在着非线性、不稳定性和多经济因素影响的复杂性,本文提出了基于支持向量机的税收预测模型,并用该模型对广东省从化市的税收预测进行了实证分析.支持向量机是统计机器学习的核心内容,它基于VC维理论和结构风险最小化理论,在模式识别、函数拟合以及回归预测等领域得到了广泛的应用.在其应用中,核函数的选择以及正则化参数的设置对支持向量机的模型有很大的影响,所以本文采用智能的算法对支持向量机中的参数自动的进行选择,经常用的智能优化算法有粒子群优化算法、遗传优化算法、网格搜索算法.本文的主要工作如下:首先,由于支持向量机中的正则化参数C和径向基核函数的参数g直接影响算法模型的预测效果,所以本文分别利用粒子群优化算法、遗传算法和网格搜索算法对正则化参数C和径向基核函数的参数g进行寻优.其次,对于影响税收收入的指标进行主成分分析,消除指标间的冗余变量,进行降维处理,把处理过的数据作为输入数据.最后,用粒子群寻优得到的最佳的C为1.1859,最佳的g为0.1;用遗传算法寻优得到的最佳的C为4.4529,最佳的g为0.018172;用网格搜索算法得到的最佳的C为16,最佳的g为0.0039063.用粒子群算法进行参数优化的支持向量机模型平均绝对百分比误差和均方百分比误差都是最小的,税收预测模型达到了很好的效果.其次是遗传算法进行优化的支持向量机模型,最后是网格搜索算法进行优化的支持向量机模型.同时本文还用未进行主成分分析的支持向量机各模型与进行主成分分析的支持向量机各模型进行对比,得到对税收数据进行主成分分析的预测结果更优,准确率更高,丰富了税收预测的研究方法.
[Abstract]:Aiming at the complexity of nonlinear, unstable and multi-economic factors in tax forecasting, this paper puts forward a tax forecasting model based on support vector machine, and makes an empirical analysis of tax forecast in Conghua City, Guangdong Province by using this model. Support vector machine (SVM) is the core of statistical machine learning. It is widely used in pattern recognition, function fitting and regression prediction based on VC dimension theory and structural risk minimization theory. In its application, the choice of kernel function and the setting of regularization parameters have great influence on the model of support vector machine, so the intelligent algorithm is used to automatically select the parameters in support vector machine. The commonly used intelligent optimization algorithms are particle swarm optimization algorithm, genetic optimization algorithm and grid search algorithm. The main work of this paper is as follows: firstly, because the regularization parameter C in support vector machine and the parameter g of radial basis function directly affect the prediction effect of the algorithm model, the particle swarm optimization algorithm is used in this paper. Genetic algorithm and mesh search algorithm are used to optimize the regularization parameter C and the parameter g of radial basis function. Secondly, the principal component analysis (PCA) is carried out on the indicators that affect the tax revenue, and the redundant variables between the indexes are eliminated, the dimensionality is reduced, and the processed data is taken as the input data. Finally, the best C is 1.1859, the best g is 0.1, the best C is 4.4529, the best g is 0.018172, and the best C is 16 and the best g is 0.0039063. The average absolute percentage error and mean square percentage error of the support vector machine model based on particle swarm optimization are minimum, and the tax forecasting model achieves good results. Secondly, the support vector machine model is optimized by genetic algorithm, and the support vector machine model is optimized by grid search algorithm. At the same time, by comparing the support vector machine models without principal component analysis and the support vector machine models of principal component analysis, the prediction results of principal component analysis for tax data are better and the accuracy is higher. It enriches the research methods of tax forecast.
【学位授予单位】：广东工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：F812.42;TP18

【参考文献】