第三届挑战赛B4-基于数据挖掘技术的市财政收入分析预测模型

全国大学生数据挖掘竞赛 优秀作品 作品名称：基于数据挖掘技术的市财政收入分析预测模型荣获奖项：一等奖作品单位：汕头大学作品成员：林西西陈炎君王莎莎指导教师：李健 基于BP神经网络的地方财政收入预测模型 本文针对广州市财政收入及影响财政收入关键因素的问题，以题目提供的各类税收收入及宏观经济和非经济指标数据为基础，利用典型相关分析、熵权系数法、灰色预测、主成分多元回归、BP神经网络预测等方法，对众多复杂的数据进行多元统计分析和预测，得到对广州市财政评价的更为深层次的探究结果。 针对问题一，通过分析原数据，可以得出了历年地区财政收入为公共财政收入与基金预算收入之和的结论，并且历年的政府性基金收益率固定，每年收入也固定，所以我们把研究影响地方财政关键因素的问题转化为研究影响公共财政收入关键宏观因素问题。 我们通过典型相关分析，即利用宏观因素和对应关联的税种收入的相关关系来衡量两组指标的关联度，得出城市居民年人均可支配收入、第二产业增加值、城市商品零售价格指数、建筑企业利润总额、第三产业增加值、住宿和餐饮业零售额、全社会房地产开发投资额、地区生产总值、批发零售业增加值以及工业增加值是影响公共财政收入的关键因素的结论，而这些因素也是影响地方财政的关键因素。另一方面，利用熵权系数模型求得与公共财政收入关联宏观因素的权重，并确定关联度较大的指标。通过比较两个模型的结论基本一致。 针对问题二，我们把财政总收入分成公共财政收入类以及基金预算收入类两类。首先，对于公共财政收入类的预测，一方面，根据影响公共财政收入的关键宏观因素，采用灰色预测模型对原始数据做累加生成得到规律性较强的近似指数序列，再对各个宏观因素作预测；另一方面，根据题目给出的历年数据，我们利用主成分回归法建立公共财政收入关于主成分的回归方程，进而预算出公共财政收入。其次，对于基金预算收入类的预测，我们采用多项式拟合的方法对历年基金预算收入拟合，并作相应预测。最后，加总公共财政收入与基金预算收入预测值得到财政总收入，我们得出2014年和2015年地方财政总收入分别为2453.9亿元和2843.6亿元。为了优化模型和克服灰色预测—主成分回归模型在处理反馈信息时的缺陷，采用BP神经网络构建地方公共财政收入预测模型，以充分挖掘公共财政收入、支出与宏观经济活动的反馈关系，最后得出2014年和2015年公共财政收入的预测值分别为1369亿元和1496.6亿元。 针对问题三，我们通过对比历年财政支出情况，给出了2015年广州市财政预算草案一些分析和建议，并提出有效支配财政收入的策略。 关键词：典型相关分析；灰色预测；BP神经网络；主成分回归分析 Local Financial Revenue Forecast Model Based on BP Neural Network AbstractAimingGuangzhou revenue and revenue key factor which affect the Government revenue problems, we base on various types of tax revenue to providethetitle and non-economic indicators and macroeconomic data.The CanonicalCorrelation Analysis,Entropy Coefficient,Gray Prediction,the main ingredientMultiple regression, BP neural network forecasting method, are used to analysis andforecast the complex data statistical tests, we get more in-depth financial evaluation ofGuangzhou exploration results. As for the question one, we can draw the calendar year revenue areas of publicrevenue and income fund budget and the conclusions of government funds throughanalying the original data , since annual income is fixed, so we change Local financialissues into finding key factors which affect public revenue study macroeconomicfactors critical issue.Byusing Canonical Correlation Analysis,using the correlation between macroeconomic factors and the corresponding revenue, to measure the correlationdegree between two sets, we conclude the key factors is as follows, urban residentspercapita disposable income,secondary industry,urban retail price index,constructionenterprises total profits,the tertiary industry,accommodation andcatering retail sales, total investment in real estate development, GDP, retail sa les, andindustrial added value. These factors are also key factors in the local financial . On theother hand, Entropy Coefficient Model is used to obtain and macroeconomic factorsassociated with heavy public revenue and identify indicators related degree. The twoconclusions are similar through comparing two models.As for the question two, we have divided the total government revenue into fund budget revenue and public income. Firstly, for the public revenue prediction, on onehand, based on the impact of public revenue according to key macroeconomic factors,the Gray Prediction Model turn the raw datato accumulated generating strongregularity approximate exponential sequence, and then make prodictions for variousmacroeconomic factors. On the other hand, based on the given historical data, we usePrincipalComponent Regression to establishment public revenue on principalcomponent regression equation, and then the budget of the public revenue. Secondly,for the kind of fund budget revenue forecast, we use polynomial fitting method to fitthe calendar year fund budget revenue, and forecast accordingly. Finally, we add thetotal public revenue and fund budget revenue forecast to total revenue worth. Thelocal fiscalrevenue forecast in2014and 2015are 245.39billion yuanand284.36billion yuan respectively.In order to overcome the gray prediction-PrincipalComponent Regression Model defects in dealing with feedback information, the useof BP neural network to build local public financial revenueforecast model, in orderto fully tap the feedback between public revenue and expenditure and macroeconomicactivity.Finally,the result of public revenue in 2014 and 2015 are 1369 billion yuanand 1496.6 billion yuan respectively.As for question three,we compare the financial expenditure over the years and give the 2015 draft budget in Guangzhou, some analysts and recommendations andpropose effective strategies disposable revenue. KeyWords:Canonical Correlation Analysis;GrayForecasting; BP neural networkPrincipal Component Regression Analysis 第页 目录 1.研究目标...............................................................................................12.分析方法与过程....................................................................................12.1.总体流程.............................................................................................................................12.2.具体步骤.......................................................................................................