CN107506852A - A kind of tax arrear Forecasting Methodology and prediction meanss based on data mining - Google Patents

A kind of tax arrear Forecasting Methodology and prediction meanss based on data mining Download PDF

Info

Publication number
CN107506852A
CN107506852A CN201710648805.0A CN201710648805A CN107506852A CN 107506852 A CN107506852 A CN 107506852A CN 201710648805 A CN201710648805 A CN 201710648805A CN 107506852 A CN107506852 A CN 107506852A
Authority
CN
China
Prior art keywords
neuron
output
input
tax
enterprise
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN201710648805.0A
Other languages
Chinese (zh)
Inventor
段然
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Foshan University
Original Assignee
Foshan University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Foshan University filed Critical Foshan University
Priority to CN201710648805.0A priority Critical patent/CN107506852A/en
Publication of CN107506852A publication Critical patent/CN107506852A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
    • G06Q40/12—Accounting
    • G06Q40/123—Tax preparation or submission

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Economics (AREA)
  • General Physics & Mathematics (AREA)
  • Finance (AREA)
  • Strategic Management (AREA)
  • Development Economics (AREA)
  • Accounting & Taxation (AREA)
  • Marketing (AREA)
  • General Business, Economics & Management (AREA)
  • Human Resources & Organizations (AREA)
  • Data Mining & Analysis (AREA)
  • Artificial Intelligence (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Computing Systems (AREA)
  • Molecular Biology (AREA)
  • General Health & Medical Sciences (AREA)
  • Evolutionary Computation (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • General Engineering & Computer Science (AREA)
  • Technology Law (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Game Theory and Decision Science (AREA)
  • Health & Medical Sciences (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Tourism & Hospitality (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses a kind of tax arrear Forecasting Methodology based on data mining, comprise the following steps, obtain the sample data of database Zhong Ge enterprises, the sample data includes multiple taxes and the taxation periods corresponding with tax；According to the data sample of each enterprise, BP neural network model is established；The sample data of each enterprise is input in BP neural network model, the BP neural network model exports the tax arrear prediction result of next taxation periods of each enterprise.The present invention passes through the existing existing government tax fillings of each enterprise of database in taxation system, establish BP neural network model, complete BP neural network model, the sample data of database Zhong Ge enterprises is input in BP neural network model again, draw the tax arrear prediction result of the next taxation periods of each enterprise, it is easy to related governmental departments to be paid close attention to according to the tax arrear prediction result of the next taxation periods of each enterprise to the enterprise that risk of owing taxes be present, effectively reduces the possibility of tax erosion caused by enterprise's tax arrear.

Description

A kind of tax arrear Forecasting Methodology and prediction meanss based on data mining
Technical field
The present invention relates to Data Mining, is more specifically to a kind of tax arrear Forecasting Methodology based on data mining Prediction meanss.
Background technology
The tax policy is an important component of state fiscal policy, and the situation of paying taxes of enterprise had both been concerned in state's family property Political affairs are taken in, while also concern the credibility in enterprise, and an enterprise, can be to country and enterprise itself if there is tax arrear situation Adversely affect, therefore country is necessary to be predicted judgement to enterprise's tax arrear situation.
But in the prior art, the tax arrear behavior that universal taxation system can not be to each enterprise is predicted, and causes nothing Method learns that enterprise illegally pays taxes behavior, and the fiscal revenues to country cause strong influence, while also result in some bad enterprises Industry behavior of illegally paying taxes is more savage.
The content of the invention
The technical problem to be solved in the present invention is：A kind of tax arrear Forecasting Methodology based on data mining is provided and predicts dress Put.
The present invention solve its technical problem solution be：
A kind of tax arrear Forecasting Methodology based on data mining, comprises the following steps：
Step A：The sample data of database Zhong Ge enterprises is obtained, the sample data includes multiple taxes and with receiving The corresponding taxation periods of the amount of tax to be paid；
Step B：According to the data sample of each enterprise, BP neural network model is established；
Step C：The sample data of each enterprise is input in BP neural network model, the BP neural network model output The tax arrear prediction result of next taxation periods of each enterprise.
As the further improvement of above-mentioned technical proposal, the BP neural network model includes multiple input neurons and defeated Go out neuron, the step B comprises the following steps：
Step B1：The connection weight between each input neuron and output neuron is initialized, sets output neuron Activation primitive, frequency of training limiting value, reference value and error precision；
Step B2：Obtain the sample data of enterprise；
Step B3：Sample data is input to the input neuron of BP neural network model, according to the defeated of input neuron Go out the output valve that value, connection weight and activation primitive calculate output neuron；
Step B4：According to the output valve and reference value of output neuron, output valve and the reference of output neuron are calculated The anticipation error of value；
Step B5：The anticipation error being calculated according to step B4, correct each input neuron and output neuron it Between connection weight；
Step B6：Whether the anticipation error that judgment step B4 is calculated is less than error precision, judges current training time Whether number is more than frequency of training limiting value, if it is desired to which error is less than error precision or frequency of training reaches the frequency of training limit Value, BP neural network model are completed；Otherwise current frequency of training adds one, obtains next sample data, return to step B3.
As the further improvement of above-mentioned technical proposal, output neuron is calculated according to formula 1 first in the step B3 Input value, formula 1 is as follows,Wherein IjRepresent the input value of output neuron, WijRepresent input Connection weight between neuron i and output neuron j, OiRepresent the output valve of input neuron；Afterwards according to activation primitive The output valve of output neuron is calculated, for the activation primitive as described in formula 2, the formula 2 is as follows,Wherein OjRepresent the output valve of output neuron.
As the further improvement of above-mentioned technical proposal, the defeated of output neuron is calculated according to formula 3 in the step B4 Go out the anticipation error of value and reference value, the formula 3 is as follows, Ej=sigmoid'(Oj)*(Tj-Oj)=Oj(1-Oj)(Tj- Oj), wherein EjRepresent the anticipation error of output neuron, TjRepresent reference value.
As the further improvement of above-mentioned technical proposal, the step B5 according to formula 4 correct it is each input neuron with Connection weight between output neuron, the formula 4 is as follows, Wij=Wij+λEjOi, wherein λ expression study rate coefficients, The learning rate coefficient value scope is between 0 to 0.1.
The beneficial effects of the invention are as follows：The present invention is existing by the existing each enterprise of database in taxation system first Government tax fillings, BP neural network model is established, complete BP neural network model, then the sample data of database Zhong Ge enterprises is defeated Enter into BP neural network model, draw the tax arrear prediction result of the next taxation periods of each enterprise, be easy to related governmental departments root The enterprise that risk of owing taxes be present is paid close attention to according to the tax arrear prediction result of the next taxation periods of each enterprise, enterprise is effectively reduced and owes The possibility of tax erosion caused by tax.
The present invention also discloses a kind of tax arrear prediction meanss based on data mining, including：
Data acquisition module, for obtaining the sample data of database Zhong Ge enterprises, the sample data includes multiple receive The amount of tax to be paid and the taxation periods corresponding with tax；
BP neural network model generation module, for the data sample according to each enterprise, establish BP neural network model；
Tax arrear prediction module, for the sample data of each enterprise to be input in BP neural network model, obtains each enterprise Next taxation periods tax arrear prediction result.
As the further improvement of above-mentioned technical proposal, the BP neural network model generation module includes：
Initialization unit, for initializing the connection weight between each input neuron and output neuron, setting is defeated Go out activation primitive, frequency of training limiting value, reference value and the error precision of neuron；
Input block：BP neural network model is input to for obtaining the sample data of enterprise, and by the sample data Input neuron；
First computing unit, output is calculated for the output valve according to input neuron, connection weight and activation primitive The output valve of neuron；
Second computing unit, for the output valve and reference value according to output neuron, calculate the defeated of output neuron Go out the anticipation error of value and reference value；
3rd computing unit, for according to anticipation error, correcting the company between each input neuron and output neuron Connect weight；
First judging unit, for judging whether anticipation error is less than error precision；
Second judging unit, for judging whether current frequency of training is more than frequency of training limiting value；
Decision package, the decision package are configured as anticipation error and reach training less than error precision or frequency of training Number limiting value, BP neural network model are completed；Otherwise current frequency of training adds one, obtains next sample data, continues to hold Row calculates.
As the further improvement of above-mentioned technical proposal, first computing unit calculates output god according to formula 1 first Input value through member, formula 1 is as follows,Wherein IjRepresent the input value of output neuron, WijTable Show the connection weight between input neuron i and output neuron j, OiRepresent the output valve of input neuron；Afterwards according to sharp Function living calculates the output valve of output neuron, and for the activation primitive as described in formula 2, the formula 2 is as follows,Wherein OjRepresent the output valve of output neuron.
As the further improvement of above-mentioned technical proposal, second computing unit calculates output neuron according to formula 3 Output valve and reference value anticipation error, the formula 3 is as follows, Ej=sigmoid'(Oj)*(Tj-Oj)=Oj(1-Oj) (Tj-Oj), wherein EjRepresent the anticipation error of output neuron, TjRepresent reference value.
As the further improvement of above-mentioned technical proposal, the 3rd computing unit corrects each input god according to formula 4 Through the connection weight between member and output neuron, the formula 4 is as follows, Wij=Wij+λEjOi, wherein λ expression learning rates Coefficient, the learning rate coefficient value scope is between 0 to 0.1.
The beneficial effects of the invention are as follows：The present invention pays taxes by the way that the existing each enterprise of database in taxation system is existing Record, BP neural network model is established, complete BP neural network model, then the sample data of database Zhong Ge enterprises is input to In BP neural network model, the tax arrear prediction result of the next taxation periods of each enterprise is drawn, is easy to related governmental departments according to each The tax arrear prediction result of the next taxation periods of enterprise is paid close attention to the enterprise that risk of owing taxes be present, and is effectively reduced enterprise's tax arrear and is drawn The possibility of the tax erosion risen.
Brief description of the drawings
Technical scheme in order to illustrate the embodiments of the present invention more clearly, make required in being described below to embodiment Accompanying drawing is briefly described.Obviously, described accompanying drawing is the part of the embodiment of the present invention, rather than is all implemented Example, those skilled in the art on the premise of not paying creative work, can also obtain other designs according to these accompanying drawings Scheme and accompanying drawing.
Fig. 1 is the tax arrear Forecasting Methodology embodiment flow chart of the present invention；
Fig. 2 is the tax arrear prediction meanss flow chart of the present invention.
Embodiment
Carried out below with reference to the design of embodiment and accompanying drawing to the present invention, concrete structure and caused technique effect clear Chu, complete description, to be completely understood by the purpose of the present invention, feature and effect.Obviously, described embodiment is this hair Bright part of the embodiment, rather than whole embodiments, based on embodiments of the invention, those skilled in the art is not paying The other embodiment obtained on the premise of creative work, belongs to the scope of protection of the invention.
1~Fig. 2 of reference picture, pay taxes feelings of the present invention for the ease of departments of government to the next taxation periods of each enterprise Condition is predicted, and the invention discloses a kind of tax arrear Forecasting Methodology based on data mining, it is characterised in that including following Step：
Step A：The sample data of database Zhong Ge enterprises is obtained, the sample data includes multiple taxes and with receiving The corresponding taxation periods of the amount of tax to be paid；
Step B：According to the data sample of each enterprise, BP neural network model is established；
Step C：The sample data of each enterprise is input in BP neural network model, the BP neural network model output The tax arrear prediction result of next taxation periods of each enterprise.
Specifically, the present invention establishes BP by the existing existing government tax fillings of each enterprise of database in taxation system Neural network model, BP neural network model is completed, then the sample data of database Zhong Ge enterprises is input to BP neural network In model, the tax arrear prediction result of the next taxation periods of each enterprise is drawn, is easy to related governmental departments to be received according to each enterprise is next The tax arrear prediction result in tax cycle is paid close attention to the enterprise that risk of owing taxes be present, and effectively reduces stream of revenue caused by enterprise's tax arrear The possibility of mistake.
It is further used as preferred embodiment, in the invention embodiment, the BP neural network model Comprise the following steps including multiple input neurons and output neuron, the step B：
Step B1：The connection weight between each input neuron and output neuron is initialized, sets output neuron Activation primitive, frequency of training limiting value, reference value and error precision；
Step B2：Obtain the sample data of enterprise；
Step B3：Sample data is input to the input neuron of BP neural network model, according to the defeated of input neuron Go out the output valve that value, connection weight and activation primitive calculate output neuron；
Step B4：According to the output valve and reference value of output neuron, output valve and the reference of output neuron are calculated The anticipation error of value；
Step B5：The anticipation error being calculated according to step B4, correct each input neuron and output neuron it Between connection weight；
Step B6：Whether the anticipation error that judgment step B4 is calculated is less than error precision, judges current training time Whether number is more than frequency of training limiting value, if it is desired to which error is less than error precision or frequency of training reaches the frequency of training limit Value, BP neural network model are completed；Otherwise current frequency of training adds one, obtains next sample data, return to step B3.
Specifically, the invention is easy to be follow-up firstly the need of some parameters being related in initialization Forecasting Methodology Calculate deterministic process and parameter is provided；Obtain the sample data of each company in database one by one afterwards, pass through multiple sample datas Continually enter, constantly train, by repeatedly to it is each input neuron and output neuron between connection weight carry out Amendment, the output valve of output neuron is set anticipation error is reduced as far as possible constantly close to reference value, ensure BP neural network mould The degree of accuracy of type.
For example illustrate herein, it is assumed that the data of paying taxes of multiple companies, each company's January are stored with database The tax of part is Z1, and the tax in 2 months is Z2, and the tax in March is Z3, and the tax in April is Z4, is then needed Predict next taxation periods, that is, the tax in May, when establishing BP neural network model, with 1,2, March pays taxes Volume is as sample data, using the tax in April as parameter value, by 1,2, the tax in March be input to input neuron In, the output valve of output neuron represents the tax in the April of BP neural network model prediction, judges BP neural network mould Difference in the tax and actual conditions in the April that type is predicted between the tax in April, as anticipation error, afterwards By the way that multiple companies are paid taxes into data input into BP neural network model, constantly to input neuron and output neuron Between connection weight be modified, when BP neural network model after accomplishing the setting up, then with each company 1,2,3, April is received The amount of tax to be paid inputs as sample data, calculates the tax of the next taxation periods of each company.
It is further used as preferred embodiment, in the invention specific embodiment, basis first in the step B3 Formula 1 calculates the input value of output neuron, and formula 1 is as follows,Wherein IjRepresent output neuron Input value, WijRepresent the connection weight between input neuron i and output neuron j, OiRepresent the output of input neuron Value；The output valve of output neuron is calculated according to activation primitive afterwards, the activation primitive is as described in formula 2, and the formula 2 is such as Shown in lower,Wherein OjRepresent the output valve of output neuron.
Preferred embodiment is further used as, in the invention embodiment, according to public affairs in the step B4 Formula 3 calculates the output valve of output neuron and the anticipation error of reference value, and the formula 3 is as follows, Ej=sigmoid' (Oj)*(Tj-Oj)=Oj(1-Oj)(Tj-Oj), wherein EjRepresent the anticipation error of output neuron, TjRepresent reference value.
It is further used as preferred embodiment, in the invention embodiment, the step B5 is according to formula 4 The connection weight between each input neuron and output neuron is corrected, the formula 4 is as follows, Wij=Wij+λEjOi, Wherein λ represents study rate coefficient, and the learning rate coefficient value scope is between 0 to 0.1.
The invention also discloses a kind of tax arrear prediction meanss based on data mining simultaneously, including：
Data acquisition module, for obtaining the sample data of database Zhong Ge enterprises, the sample data includes multiple receive The amount of tax to be paid and the taxation periods corresponding with tax；
BP neural network model generation module, for the data sample according to each enterprise, establish BP neural network model；
Tax arrear prediction module, for the sample data of each enterprise to be input in BP neural network model, obtains each enterprise Next taxation periods tax arrear prediction result.
It is further used as preferred embodiment, in the invention embodiment, the BP neural network model Generation module includes：
Initialization unit, for initializing the connection weight between each input neuron and output neuron, setting is defeated Go out activation primitive, frequency of training limiting value, reference value and the error precision of neuron；
Input block：BP neural network model is input to for obtaining the sample data of enterprise, and by the sample data Input neuron；
First computing unit, output is calculated for the output valve according to input neuron, connection weight and activation primitive The output valve of neuron；
Second computing unit, for the output valve and reference value according to output neuron, calculate the defeated of output neuron Go out the anticipation error of value and reference value；
3rd computing unit, for according to anticipation error, correcting the company between each input neuron and output neuron Connect weight；
First judging unit, for judging whether anticipation error is less than error precision；
Second judging unit, for judging whether current frequency of training is more than frequency of training limiting value；
Decision package, the decision package are configured as anticipation error and reach training less than error precision or frequency of training Number limiting value, BP neural network model are completed；Otherwise current frequency of training adds one, obtains next sample data, continues to hold Row calculates.
It is further used as preferred embodiment, in the invention embodiment, first computing unit is first The input value of output neuron is first calculated according to formula 1, formula 1 is as follows,Wherein IjRepresent defeated Go out the input value of neuron, WijRepresent the connection weight between input neuron i and output neuron j, OiRepresent input nerve The output valve of member；The output valve of output neuron is calculated according to activation primitive afterwards, the activation primitive is as described in formula 2, institute It is as follows to state formula 2,Wherein OjRepresent the output valve of output neuron.
It is further used as preferred embodiment, in the invention embodiment, the second computing unit root The output valve of output neuron and the anticipation error of reference value are calculated according to formula 3, the formula 3 is as follows, Ej= sigmoid'(Oj)*(Tj-Oj)=Oj(1-Oj)(Tj-Oj), wherein EjRepresent the anticipation error of output neuron, TjRepresent reference Value.
Preferred embodiment is further used as, the 3rd computing unit corrects each input neuron according to formula 4 Connection weight between output neuron, the formula 4 is as follows, Wij=Wij+λEjOi, wherein λ expression learning rates system Number, the learning rate coefficient value scope is between 0 to 0.1.
The better embodiment of the present invention is illustrated above, but the invention is not limited to the implementation Example, those skilled in the art can also make a variety of equivalent modifications on the premise of without prejudice to spirit of the invention or replace Change, these equivalent modifications or replacement are all contained in the application claim limited range.

Claims (10)

1. a kind of tax arrear Forecasting Methodology based on data mining, it is characterised in that comprise the following steps：
Step A：The sample data of database Zhong Ge enterprises is obtained, the sample data includes multiple taxes and and tax Corresponding taxation periods；
Step B：According to the data sample of each enterprise, BP neural network model is established；
Step C：The sample data of each enterprise is input in BP neural network model, the BP neural network model exports each enterprise The tax arrear prediction result of next taxation periods of industry.
A kind of 2. tax arrear Forecasting Methodology based on data mining according to claim 1, it is characterised in that：The BP nerves Network model includes multiple input neurons and output neuron, the step B and comprised the following steps：
Step B1：The connection weight between each input neuron and output neuron is initialized, sets swashing for output neuron Function, frequency of training limiting value, reference value and error precision living；
Step B2：Obtain the sample data of enterprise；
Step B3：Sample data is input to the input neuron of BP neural network model, according to input neuron output valve, Connection weight and activation primitive calculate the output valve of output neuron；
Step B4：According to the output valve and reference value of output neuron, the output valve and reference value of output neuron are calculated Anticipation error；
Step B5：The anticipation error being calculated according to step B4, correct between each input neuron and output neuron Connection weight；
Step B6：Whether the anticipation error that judgment step B4 is calculated is less than error precision, judges that current frequency of training is It is no to be more than frequency of training limiting value, if it is desired to error is less than error precision or frequency of training reaches frequency of training limiting value, BP neural network model is completed；Otherwise current frequency of training adds one, obtains next sample data, return to step B3.
A kind of 3. tax arrear Forecasting Methodology based on data mining according to claim 2, it is characterised in that：The step B3 In first according to formula 1 calculate output neuron input value, formula 1 is as follows,Wherein IjTable Show the input value of output neuron, WijRepresent the connection weight between input neuron i and output neuron j, OiRepresent input The output valve of neuron；The output valve of output neuron, the activation primitive such as institute of formula 2 are calculated according to activation primitive afterwards To state, the formula 2 is as follows,Wherein OjRepresent the defeated of output neuron Go out value.
A kind of 4. tax arrear Forecasting Methodology based on data mining according to claim 3, it is characterised in that：The step B4 The anticipation error of the middle output valve and reference value that output neuron is calculated according to formula 3, the formula 3 is as follows, Ej= sigmoid'(Oj)*(Tj-Oj)=Oj(1-Oj)(Tj-Oj), wherein EjRepresent the anticipation error of output neuron, TjRepresent reference Value.
A kind of 5. tax arrear Forecasting Methodology based on data mining according to claim 4, it is characterised in that：The step B5 Connection weight between each input neuron and output neuron is corrected according to formula 4, the formula 4 is as follows, Wij= Wij+λEjOi, wherein λ expression study rate coefficients, the learning rate coefficient value scope is between 0 to 0.1.
A kind of 6. tax arrear prediction meanss based on data mining, it is characterised in that including：
Data acquisition module, for obtaining the sample data of database Zhong Ge enterprises, the sample data includes multiple taxes And the taxation periods corresponding with tax；
BP neural network model generation module, for the data sample according to each enterprise, establish BP neural network model；
Tax arrear prediction module, for the sample data of each enterprise to be input in BP neural network model, is obtained under each enterprise The tax arrear prediction result of one taxation periods.
A kind of 7. tax arrear prediction meanss based on data mining according to claim 6, it is characterised in that the BP nerves Network model generation module includes：
Initialization unit, for initializing the connection weight between each input neuron and output neuron, setting output god Activation primitive, frequency of training limiting value, reference value and error precision through member；
Input block：BP neural network mode input is input to for obtaining the sample data of enterprise, and by the sample data Neuron；
First computing unit, output nerve is calculated for the output valve according to input neuron, connection weight and activation primitive The output valve of member；
Second computing unit, for the output valve and reference value according to output neuron, calculate the output valve of output neuron With the anticipation error of reference value；
3rd computing unit, for according to anticipation error, correcting the connection weight between each input neuron and output neuron Weight；
First judging unit, for judging whether anticipation error is less than error precision；
Second judging unit, for judging whether current frequency of training is more than frequency of training limiting value；Decision package, it is described to determine Plan unit is configured as anticipation error and reaches frequency of training limiting value, BP neural network mould less than error precision or frequency of training Type is completed；Otherwise current frequency of training adds one, obtains next sample data, continues executing with calculating.
A kind of 8. tax arrear prediction meanss based on data mining according to claim 7, it is characterised in that：First meter The input value that unit calculates output neuron according to formula 1 first is calculated, formula 1 is as follows,Wherein Ij Represent the input value of output neuron, WijRepresent the connection weight between input neuron i and output neuron j, OiRepresent defeated Enter the output valve of neuron；The output valve of output neuron, the activation primitive such as institute of formula 2 are calculated according to activation primitive afterwards To state, the formula 2 is as follows,Wherein OjRepresent the output of output neuron Value.
A kind of 9. tax arrear prediction meanss based on data mining according to claim 8, it is characterised in that：Second meter Calculate unit and the output valve of output neuron and the anticipation error of reference value are calculated according to formula 3, the formula 3 is as follows, Ej =sigmoid'(Oj)*(Tj-Oj)=Oj(1-Oj)(Tj-Oj), wherein EjRepresent the anticipation error of output neuron, TjRepresent ginseng Examine value.
A kind of 10. tax arrear prediction meanss based on data mining according to claim 9, it is characterised in that：Described 3rd Computing unit corrects the connection weight between each input neuron and output neuron, the following institute of formula 4 according to formula 4 Show, Wij=Wij+λEjOi, wherein λ expression study rate coefficients, the learning rate coefficient value scope is between 0 to 0.1.
CN201710648805.0A 2017-08-01 2017-08-01 A kind of tax arrear Forecasting Methodology and prediction meanss based on data mining Pending CN107506852A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201710648805.0A CN107506852A (en) 2017-08-01 2017-08-01 A kind of tax arrear Forecasting Methodology and prediction meanss based on data mining

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201710648805.0A CN107506852A (en) 2017-08-01 2017-08-01 A kind of tax arrear Forecasting Methodology and prediction meanss based on data mining

Publications (1)

Publication Number Publication Date
CN107506852A true CN107506852A (en) 2017-12-22

Family

ID=60689776

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201710648805.0A Pending CN107506852A (en) 2017-08-01 2017-08-01 A kind of tax arrear Forecasting Methodology and prediction meanss based on data mining

Country Status (1)

Country Link
CN (1) CN107506852A (en)

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102622418A (en) * 2012-02-21 2012-08-01 北京联合大学 Prediction device and equipment based on BP (Back Propagation) nerve network
CN106529729A (en) * 2016-11-18 2017-03-22 同济大学 Method and system for forecasting default of credit card user based on BP_Adaboost model

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102622418A (en) * 2012-02-21 2012-08-01 北京联合大学 Prediction device and equipment based on BP (Back Propagation) nerve network
CN106529729A (en) * 2016-11-18 2017-03-22 同济大学 Method and system for forecasting default of credit card user based on BP_Adaboost model

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
时待吾: "基于数据挖掘的企业欠税预测研究", 《中国优秀硕士学位论文全文数据库》 *

Similar Documents

Publication Publication Date Title
Challoumis-Κωνσταντίνος Χαλλουμής Arm's Length Principle and Fix Length Principle Mathematical Approach
Lin Pricing VIX futures: Evidence from integrated physical and risk‐neutral probability measures
Sanjuán‐López et al. Volatility effects of index trading and spillovers on US agricultural futures markets: A multivariate GARCH approach
CN109086954B (en) Prediction method, device, equipment and medium for predicting yield based on fund flow
CN115577851A (en) Energy consumption prediction method, device, equipment and storage medium
CN113093537B (en) Event-triggered observer design method based on online asynchronous front-part reconstruction
Kumar et al. Nonlinear effect of air travel tourism demand on economic growth in Fiji
CN107506852A (en) A kind of tax arrear Forecasting Methodology and prediction meanss based on data mining
Xie et al. Do macroprudential policies reduce risk spillovers between energy markets?: Evidence from time-frequency domain and mixed-frequency methods
Pirovano et al. Eurozone inflation differentials and the ECB
Parail Properties of electricity prices and the drivers of interconnector revenue
Perovic et al. Government expenditures composition and growth in EU15: a dynamic heterogeneous approach
Lou Liability-Side Pricing of Swaps and Coherent CVA and FVA by Regression/Simulation
Steele A NOTE ON ESTIMATING THE INTERNAL RATE OF RETURN FROM PUBLISHED FINANCIAL STATEMENTS.
Loeser Consumer Surplus with Incomplete Markets: Applications to Savings and Microfinance
Han et al. Affine models of the joint dynamics of exchange rates and interest rates
CN113592366A (en) Business strategy making method and device based on artificial intelligence
CN113450227A (en) Self-adaptive stock trading method based on SAC reinforcement learning
CN108734461A (en) The processing method and terminal device of pre-payment virtual resource
Lee et al. Do leveraged/inverse ETFs wag the underlying market? Evidence from the Korean stock market
Pereiro Prototyping Growth in Technology Ventures: A Practical Approach
Maria Integrated support system of process management of innovative projects
Melkonyan et al. Food safety regulations under ambiguity
Agostino et al. Recent Developments in FATCA Compliance
Sulaimani Analysis of container throughput: Demand forecast and seaport competitiveness assessment

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
RJ01 Rejection of invention patent application after publication
RJ01 Rejection of invention patent application after publication

Application publication date: 20171222