CN110223164A - Air control method and system based on transfer learning, computer installation, storage medium - Google Patents
Air control method and system based on transfer learning, computer installation, storage medium Download PDFInfo
- Publication number
- CN110223164A CN110223164A CN201910498007.3A CN201910498007A CN110223164A CN 110223164 A CN110223164 A CN 110223164A CN 201910498007 A CN201910498007 A CN 201910498007A CN 110223164 A CN110223164 A CN 110223164A
- Authority
- CN
- China
- Prior art keywords
- sample
- weight
- target
- source
- iteration
- 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
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
- G06F18/24155—Bayesian classification
-
- 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
- G06Q40/03—Credit; Loans; Processing thereof
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Finance (AREA)
- Data Mining & Analysis (AREA)
- Accounting & Taxation (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Probability & Statistics with Applications (AREA)
- Development Economics (AREA)
- Economics (AREA)
- Marketing (AREA)
- Strategic Management (AREA)
- Technology Law (AREA)
- General Business, Economics & Management (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The present invention provides a kind of air control method based on transfer learning, the method screens several samples similar with target sample in the sample of source, source sample and target sample training Weak Classifier after combined screening, combination Weak Classifier generates strong classifier to generate risk identification model, it identifies that pending cash borrows the risk class of business according to the information that pending cash borrows business using risk identification model generated, and air control is carried out according to recognition result.The present invention also provides system, the computer installations, storage medium of realizing the air control method based on transfer learning.The present invention can borrow business to cash and carry out effective risk profile.
Description
Technical field
The present invention relates to field of computer technology, and in particular to it is a kind of by the air control method and system of transfer learning, based on
Calculation machine device, storage medium.
Background technique
In the business of business bank, common credit is borrowed and cash loan is the common pattern of lending of two classes.It is seen on surface, it is general
Communication is roughly the same with the user's material borrowed with cash borrows audit, but due to risk point difference, leads to air control requirement also difference
It is very big.Common credit loan is exactly borrower without guarantee or third party guarantees, is only used as and refunds by borrower's creditworthiness
Guarantee.The main body for carrying out normal credit requirement business is business bank, finance company, electronic banking mechanism etc..General letter by ordinary
It is 1-3 with the length of maturity, the amount of the loan is thousands of to tens of thousands of.The risk of fiduciary loan is larger, needs the operation to borrower
Situation, economic benefit, development prospect comprehensive consideration.Cash loan is a kind of small amount short-term loan product, has the characteristics that following five:
Amount is small, the period is short, process is fast, high-efficient, and the length of maturity is generally 7-30 days, and the average amount of the loan is 1000, year interest rate
Averagely between 50%-200%.Type of business is borrowed according to cash currently on the market, can be divided by means of with also class loan, hair
Provide a loan firewood day (payday loan), short term loan, four major class of medium term loan.It can be seen that the debt-credit threshold that cash is borrowed is low, it is
A kind of profit mode of dear money high risk, the air control borrowed to cash are particularly important.
Summary of the invention
In view of the foregoing, it is necessary to propose a kind of air control method and system based on transfer learning, computer installation, deposit
Storage media, to realize the effective air control borrowed to cash.
The first aspect of the present invention provides a kind of air control method based on transfer learning, which comprises
Obtaining step obtains source sample D 'sWith target sample Dt, wherein the source sample D 'sIt is described including k sample
Target sample DtIncluding m sample, by the source sample D 'sWith target sample DtAs total sample, thus the number etc. of total sample
In the sum of the number k and the number m of the target sample of the source sample；
Initial weight is respectively set for each sample in total sample in first setting steps, and according to the gross sample
The weight of each sample in this obtains the weight vectors of each sample；
Second setting steps, setting maximum number of iterations are n times, and t indicates the number of iterations；
First executes step, the weight vectors of weight and every sample based on each sample in total sample
Calculate weight ratio of each sample in the t times iteration；
Second executes step, by the source sample D 'sWith the target sample DtMixing obtains mixed total sample, with
Weight of the linear regression as base classifier, by mixed total sample in conjunction with each sample in the t times iteration
The ratio training base classifier obtains a Weak Classifier；
Third executes step, calculates target sample D described in the Weak Classifier misclassificationtError rate；
4th executes step, calculates the source sample D 'sWeight renewal rate, and according to the target sample DtQuilt
The error rate of misclassification calculates the target sample DtWeight renewal rate；
5th executes step, according to the target sample DtWeight renewal rate and the source sample D 'sWeight more
New rate determines the source sample D 'sWith target sample DtIn weight of each sample in the t+1 times iteration；
Whether judgment step, judgement currently have been carried out n times iteration, wherein if currently having not carried out n times iteration,
Described first, which is triggered, after setting t=t+1 executes step；If currently having been carried out n times iteration, triggering the 6th executes step；
Described 6th executes step, according to the Weak Classifier difference for training acquisition during the N/2 times to iv-th iteration
Prediction to the risk class of total sample combines this N/2 times and trains the weak typing obtained in the process to iv-th iteration
Device obtains a strong classifier；And
7th executes step, using the strong classifier as risk identification model, using the risk identification model according to
The information of pending loan transaction identifies the risk class of pending loan transaction, and carries out air control according to recognition result.
Preferably, the target sample DtThe included m sample is that cash borrows sample, the source sample D 'sIt is wrapped
The k sample included is normal credit requirement sample, and this method obtains the source sample D 'sThe step of include:
Obtain sample set Ds, wherein the sample set DsSample including n normal credit requirement；
Each sample in sample that sample and the m cash to the n normal credit requirement are borrowed is marked,
Include: using the n normal credit requirement sample as negative sample, is each sample in the n normal credit requirement sample
One public field Sample_Flag of this increase, and Sample_Flag=-1 is set；The m cash is borrowed into sample as just
Sample borrows each sample in sample for the m cash and increases a public field Sample_Flag, and Sample_ is arranged
Flag=1；
The n normal credit requirement sample after label is borrowed sample with the m cash to mix, by mixed institute
Having the random cutting of sample is training set and verifying collection；
Using naive Bayesian (NB) as base classifier, with the public field of each sample included by the training set
The value of Sample_Flag is as the tag along sort training base classifier；
The test set is predicted using the trained base classifier, exports each sample in the test set
This is the probability of positive sample；
According to the size for the probability that each sample exported is positive sample to all samples included by the test set
Descending arrangement is carried out, the preceding k normal credit requirement sample after filtering out descending arrangement, by the k filtered out common credits
Sample provide a loan as the source sample D 's。
Preferably, weight ratio of each sample in total sample in the t times iteration is by formula:It is calculated, whereinRepresent power of each sample in the t times iteration in total sample
Weight ratio,Weight vectors of each sample in the t times iteration in total sample are represented,It represents in total sample
Weight of each sample in the t times iteration.
Preferably, in the second execution step: setting tag along sort 1 represents high default risk, wind of breaking a contract in 0 representative
Danger, -1 represents low default risk, and remembers the Weak Classifier that the training obtains are as follows: ht: X → Y, Y ∈ { -1,0,1 }；Wherein, this is weak
Classifier htBy the Probability p of each sample classification in the softmax layers of output k+m sample to each tag along sort, 0
≤p≤1。
Preferably, the error rate is according to formulaIt is calculated, wherein ∈tGeneration
Error rate described in table, | ht(xi)-c(xi) | indicate that the Weak Classifier exports the target sample DtIn each sample be true
The probability of tag along sort value and 1 difference, it is describedRepresent the target sample DtIn each sample in the t times iteration
Weight.
Preferably, the target sample DtWeight renewal rateThe source sample D 'sWeight update speed
RateWherein, n=k, N are the set maximum number of iterations.
Preferably, weight of each sample in the t+1 times iteration are as follows:
Wherein,Weight of each sample in the t times iteration is represented,Each sample is then represented at the t+1 times
Weight when iteration, | ht(xi)-c(xi) | indicate that the Weak Classifier exports the target sample DtIn each sample be true
The probability of tag along sort value and 1 difference.
Second aspect of the present invention provides a kind of computer installation, and the computer installation includes memory and processor, institute
Memory is stated for storing at least one instruction, the processor is described based on migration for realizing when at least one described instruction
The air control method of study.
Third aspect present invention provides a kind of computer readable storage medium, and the computer-readable recording medium storage has
At least one instruction, at least one described instruction realize the air control method based on transfer learning when being executed by processor.
Fourth aspect present invention provides a kind of air control system based on transfer learning, runs on computer installation, the system
System includes:
Module is obtained, for obtaining source sample D 'sWith target sample Dt, wherein the source sample D 'sIncluding k sample,
The target sample DtIncluding m sample, by the source sample D 'sWith target sample DtAs total sample, thus the number of total sample
Mesh is equal to the sum of the number k and the number m of the target sample of the source sample；
Execution module, for initial weight to be respectively set for each sample in total sample, and according to the gross sample
The weight of each sample in this obtains the weight vectors of each sample；
The execution module, being also used to be arranged maximum number of iterations is n times, and t indicates the number of iterations；
The execution module is also used to weight based on each sample in total sample and each sample
Weight ratio of each sample described in weight vector computation in the t times iteration；
The execution module is also used to the source sample D 'sWith the target sample DtMixing obtains mixed gross sample
This, the weight using linear regression as base classifier, by mixed total sample in conjunction with each sample in the t times iteration
The ratio training base classifier obtains a Weak Classifier；
The execution module is also used to calculate target sample D described in the Weak Classifier misclassificationtError rate；
The execution module is also used to calculate the source sample D 'sWeight renewal rate, and according to the target sample
This DtThe target sample D is calculated by the error rate of misclassificationtWeight renewal rate；
The execution module is also used to according to the target sample DtWeight renewal rate and the source sample D 's's
Weight renewal rate determines the source sample D 'sWith target sample DtIn weight of each sample in the t+1 times iteration；
The execution module is also used to currently whether have been carried out n times iteration, wherein change if currently having not carried out n times
In generation, triggers the first execution step after t=t+1 is then arranged；If currently have been carried out n times iteration, triggering the 6th is held
Row step；
The execution module is also used to according to the Weak Classifier point for training acquisition during the N/2 times to iv-th iteration
The prediction of the other risk class to total sample combines this N/2 times and trains weak point obtained in the process to iv-th iteration
Class device obtains a strong classifier；And
The execution module is also used to utilize the risk identification mould using the strong classifier as risk identification model
Type identifies the risk class of pending loan transaction according to the information of pending loan transaction, and is carried out according to recognition result
Air control.
Air control method and system described in the embodiment of the present invention based on transfer learning, computer installation, storage medium,
Cash can be borrowed and carry out effective air control.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
The embodiment of invention for those of ordinary skill in the art without creative efforts, can also basis
The attached drawing of offer obtains other attached drawings.
Fig. 1 is the flow chart for the air control method based on transfer learning that present pre-ferred embodiments provide.
Fig. 2 is the module map for the air control system based on transfer learning that present pre-ferred embodiments provide.
Fig. 3 is the schematic diagram for the computer installation that present pre-ferred embodiments provide.
The present invention that the following detailed description will be further explained with reference to the above drawings.
Specific embodiment
To better understand the objects, features and advantages of the present invention, with reference to the accompanying drawing and specific real
Applying example, the present invention will be described in detail.It should be noted that in the absence of conflict, the embodiment of the present invention and embodiment
In feature can be combined with each other.
In the following description, numerous specific details are set forth in order to facilitate a full understanding of the present invention, described embodiment is only
It is only a part of the embodiment of the present invention, instead of all the embodiments.Based on the embodiments of the present invention, ordinary skill
Personnel's every other embodiment obtained without making creative work, shall fall within the protection scope of the present invention.
Unless otherwise defined, all technical and scientific terms used herein and belong to technical field of the invention
The normally understood meaning of technical staff is identical.Term as used herein in the specification of the present invention is intended merely to description tool
The purpose of the embodiment of body, it is not intended that in the limitation present invention.
Fig. 1 is the flow chart for the air control method based on transfer learning that present pre-ferred embodiments provide.
In the present embodiment, the air control method based on transfer learning can be applied in computer installation, for needing
The computer installation of the air control based on transfer learning is carried out, can directly integrate method institute of the invention on a computing means
What is provided is used for the wind control function based on transfer learning, or with Software Development Kit (Software Development
Kit, SDK) form operation on a computing means.
As shown in Figure 1, the air control method based on transfer learning specifically includes following steps, according to different requirements,
The sequence of step can change in the flow chart, and certain steps can be omitted.
Computer installation is filtered out from the sample that a large amount of general cash is provided a loan and the m first with SPY algorithm
Cash borrows the similar k general cash loan sample of sample.Specifically comprise the following steps S11-S15.
Step S11, computer installation obtains the sample set D of normal credit requirementsThe sample set D borrowed with casht, wherein institute
State sample set DsIncluding n normal credit requirement sample, the sample set DtSample is borrowed including m cash.
In the present embodiment, each sample in the n normal credit requirement sample includes, but are not limited to creditor's
Age, educational background, annual income, the amount of the loan, the intended use of the loan, the length of maturity.The m cash borrows each sample packet in sample
It includes, but is not limited to, age, educational background, annual income, the amount of the loan, the intended use of the loan, the length of maturity of creditor.
In the present embodiment, n and m are positive integer, and n is greater than m.
In one embodiment, described in computer installation can be obtained from the server (not shown) of communication connection
N normal credit requirement sample and the m cash borrow sample.It is data such as one relevant that the server is stored with loan
Or age, the educational background, annual income, the amount of the loan, loan of the credit information such as creditor of all previous loan that is handled of more banks
Purposes, length of maturity etc..
Step S12, in the sample that computer installation borrows the sample of the n normal credit requirement and the m cash
Each sample be marked, comprising: be the n common credits using the n normal credit requirement sample as negative sample
Each sample in loan sample increases a public field Sample_Flag, and Sample_Flag=-1 is arranged；By the m
A cash borrows sample as positive sample, borrows each sample in sample for the m cash and increases a public field Sample_
Flag, and Sample_Flag=1 is set.
Step S13, the n normal credit requirement sample after label is borrowed sample with the m cash by computer installation
Mixed all random cuttings of sample are that training set and verifying collect, wherein the training set accounts for all samples by this mixing
First preset ratio (such as 70%), the verifying collection account for the second preset ratio (such as 30%) of all samples.
Step S14, computer installation is using naive Bayesian (NB) as base classifier, included by the training set
The value (i.e. 1 and -1) of the public field Sample_Flag of each sample is as the tag along sort training base classifier.
Step S15, computer installation utilizes the trained base classifier, predicts the test set, defeated
Each sample is that the probability of positive sample (that is to say that the Sample_Flag=1's for exporting each sample is general in the test set out
Rate).Computer installation according to each sample exported for the probability of positive sample size to institute included by the test set
There is sample to carry out descending arrangement, the preceding k normal credit requirement sample after filtering out descending arrangement is general by the k filtered out
Lead to fiduciary loan sample as sample set D 's.It has been filtered out from the n normal credit requirement sample as a result, and the m
The most similar k normal credit requirement sample of sample that cash is borrowed.
In one embodiment, k is less than or equal to the positive integer of m.Preferably, k is equal to m.
It should be noted that following description can be by the sample set D ' for convenience of the clear simple description present inventions
The included k normal credit requirement sample and the sample set DtThe included m cash is borrowed sample and is referred to as
" k+m sample ".Using the k+m sample as total sample.
Step S16, computer installation is that initial weight is respectively set in each sample in the k+m sample.Computer dress
Set the weight vectors that each sample is obtained according to the weight of each sample in the k+m sample.
In the present embodiment, the initial weight that each sample is arranged in computer installation is 1.That is w1=w2=wk+m=1.The k+
The weight vectors of m sample
Step S17, computer installation setting maximum number of iterations is n times, and t indicates the number of iterations.Therefore, [1, N] t ∈.
When iteration for the first time, the value of t is equal to 1.
Step S18, weight and each sample of the computer installation based on each sample in the k+m sample
Weight vector computation described in weight ratio of each sample in the t times iteration.
In the present embodiment, weight ratio of each sample in the t times iteration is by formula: It is calculated.Wherein,Represent weight of each sample in the t times iteration in the k+m sample
Ratio,Represent weight vectors of each sample in the t times iteration in the k+m sample.Represent the k+m sample
Weight of each sample in the t times iteration in this.
For in other words, weight ratio of each sample in the t times iteration is equal to the k+m in the k+m sample
In a sample weight vectors of each sample in the t times iteration divided by the k+m sample the sum of weight.
Step S19, computer installation is by the sample set D 'sAs source sample, and by the sample set DtAs target
Sample, by the source sample D 'sWith target sample DtMixing (that is to say the k normal credit requirement sample and the m
Cash borrows sample mixing).Computer installation is using linear regression (LR) as base classifier, by the mixed k+m sample
The base classifier is trained to instruct in conjunction with weight ratio of each sample in the t times iteration in the k+m sample
Practice and obtains a Weak Classifier.
In the present embodiment, computer installation setting tag along sort 1 represents high default risk, default risk in 0 representative, -1 generation
The low default risk of table (i.e. different classification values represents different default risk degree), and remember the Weak Classifier that the training obtains
Are as follows: ht: X → Y, Y ∈ { -1,0,1 }.
In the present embodiment, Weak Classifier htPass through each sample classification in the softmax layers of output k+m sample
To the Probability p of each tag along sort (i.e. 1,0, and -1), 0≤p≤1.
Step S20, computer installation calculates target sample D described in the Weak Classifier misclassificationtError rate.
Wherein, error rateWherein, | ht(xi)-c(xi) | indicate described weak point
Class device exports the target sample DtIn each sample be true tag along sort value probability and 1 difference.In other words ht(xi)
It represents the Weak Classifier and exports the target sample D in the t times iterationtIn each sample be true tag along sort value it is general
Rate, c (xi) value be considered as 1.It is describedRepresent the target sample DtIn weight of each sample in the t times iteration.
Step S21, computer installation is according to the target sample DtThe target sample D is calculated by the error rate of misclassificationt
Weight renewal rate.Computer installation also calculates the source sample D 'sWeight renewal rate.
In the present embodiment, the target sample DtWeight renewal rateThe source sample D 'sWeight
Renewal rateWherein, n=k (that is to say the source sample D 'sIncluded common credit
The quantity of loan sample), N is the set maximum number of iterations.
Step S22, computer installation is according to the target sample DtWeight renewal rate and the source sample D 's's
Weight renewal rate determines the source sample D 'sWith target sample DtIn weight of each sample in the t+1 times iteration.
Specifically, weight of each sample in the t+1 times iteration are as follows:
Wherein,Weight of each sample in the t times iteration is represented,Each sample is then represented at the t+1 times
Weight when iteration.|ht(xi)-c(xi) | indicate that the Weak Classifier exports the target sample DtIn each sample be true
The probability of tag along sort value and 1 difference.
It can be seen that for the target sample DtIn sample improve the target sample D if classification errort
In weighted value of the sample in the t+1 times iteration, so that Weak Classifier is paid close attention to the sample for being classified mistake.And for described
Source sample D 'sIn sample reduce the source sample D ' if classification errorsIn power of the sample in the t+1 times iteration
Weight values, this is because classification error is considered as this partial data and target sample differs greatly.
Step S23, computer installation judgement currently whether have been carried out n times iteration (that is to say the current t of comparison value and
The size relation of the value of N).If currently having not carried out n times iteration (i.e. current t < N), step S18 is returned to after t=t+1 is set,
To carry out next iteration.If currently having been carried out n times iteration (i.e. current t=N), S24 is thened follow the steps.
Step S24, computer installation is according to the Weak Classifier point for training acquisition during the N/2 times to iv-th iteration
(that is to say each sample classification of prediction arrives the general of each tag along sort for the prediction of the other risk class to the k+m sample
Rate), it combines this N/2 times and trains the Weak Classifier obtained in the process to iv-th iteration, obtain a strong classifier.
For example, the strong classifier
The first condition refers to: the Weak Classifier that training obtains during described the N/2 times to iv-th iteration predicts institute
It states k+m sample classification and is greater than to the sum of the probability of tag along sort 1 and predict the k+m sample classification to the general of tag along sort 0
The sum of rate and it is greater than the sum of the probability value for predicting the k+m sample classification to tag along sort -1.The second condition refers to: institute
State the N/2 times predicts the k+m sample classification to tag along sort 0 to the Weak Classifier that training during iv-th iteration obtains
The sum of probability be greater than and predict described k+m sample classification to the sum of the probability of tag along sort 1 and greater than the prediction k+m sample
Originally it is categorized into the sum of the probability value of tag along sort -1.The third condition refers to: during described the N/2 times to iv-th iteration
The Weak Classifier that training obtains predicts the k+m sample classification to the sum of the probability of tag along sort -1 k+m described greater than prediction
A sample classification is to the sum of the probability of tag along sort 1 and is greater than the probability value for predicting the k+m sample classification to tag along sort 0
The sum of.
According to above-mentioned steps S17-S24 it is found that computer installation can borrow sample and utilization based on the cash obtained in advance
The normal credit requirement sample that SPY algorithm is filtered out obtains carrying out risk for borrowing cash by TrAdaBoost training
The classifier of identification.
Step S25, computer installation utilizes the risk identification model using the strong classifier as risk identification model
The risk class of pending loan transaction is identified according to the information of pending loan transaction, and wind is carried out according to recognition result
Control.
In the present embodiment, the pending loan transaction is that cash borrows business.The letter of the pending loan transaction
Breath includes, but are not limited to age, educational background, annual income, the amount of the loan, the intended use of the loan, the length of maturity of creditor.
For example, when predicting that a certain pending loan transaction is categorized into the probability of high default risk using the strong classifier
When greater than the probability for being categorized into middle default risk and being greater than the probability for being categorized into low default risk, computer installation determines that this is a certain
Loan transaction is high default risk business, and issues and be prompted to business personnel.
In one embodiment, when determining a certain loan transaction is high default risk business, computer installation may be used also
To determine the key message for influencing the risk class of a certain loan transaction using the risk identification model.Computer installation is also
The key message is prompted to business personnel, so that business personnel can be using these key messages as the ginseng communicated with client
Examine information.
In one embodiment, described that the levels of risk for influencing a certain loan transaction is determined using the risk identification model
Other key message includes step (a1)-(a3):
(a1) information of a certain loan transaction is modified.
It for example, can be in the premise for keeping creditor's age, educational background, annual income, the intended use of the loan, the loan time limit constant
The lower modification amount of the loan such as reduces the amount of the loan.
(a2) it is identified using the risk identification model according to the information of a certain loan transaction described after modification modified
The risk class of a certain loan transaction.
(a3) when the risk class of the modified a certain loan transaction is lower than a certain loan transaction before modifying
When risk class, using the information modified as the key message for the risk class for influencing a certain loan transaction.
In conclusion the air control method described in the embodiment of the present invention based on transfer learning, first with SPY algorithm from
K sample similar with m target sample is filtered out in a large amount of normal credit requirement sample, is then instructed by TrAdaBoost
It gets for borrowing the risk identification model for carrying out risk identification to cash, business can be borrowed to cash and carry out effective risk identification
And control.
Above-mentioned Fig. 1 describes the air control method of the invention based on transfer learning in detail, below with reference to Fig. 2 and Fig. 3, to reality
The functional module and the realization wind based on transfer learning of the software systems of the existing air control method based on transfer learning
The hardware device framework of prosecutor method is introduced.
It should be appreciated that the embodiment is only purposes of discussion, do not limited by this structure in patent claim.
As shown in fig.2, being the structure chart for the air control system based on transfer learning that present pre-ferred embodiments provide.
In some embodiments, the air control system 30 based on transfer learning is run in computer installation.The meter
Calculation machine device is by being connected to the network computer installation.The air control system 30 based on transfer learning may include multiple by journey
Functional module composed by sequence code segment.The program code of each program segment in the air control system 30 based on transfer learning
It can store in the memory of computer installation, and as performed by least one described processor, (be detailed in Fig. 2 to retouch with realization
State) air control based on transfer learning.
In the present embodiment, the function of the air control system 30 according to performed by it based on transfer learning can be divided
For multiple functional modules.The functional module may include: to obtain module 301, execution module 302.The so-called module of the present invention
Refer to it is a kind of performed by least one processor and can complete the series of computation machine program segment of fixed function,
Storage is in memory.In the present embodiment, it will be described in detail in subsequent embodiment about the function of each module.
Obtain the sample set D that module 301 obtains normal credit requirementsThe sample set D borrowed with casht, wherein the sample
Collect DsIncluding n normal credit requirement sample, the sample set DtSample is borrowed including m cash.In the present embodiment, the n general
Each sample in logical fiduciary loan sample includes, but are not limited to the age of creditor, educational background, annual income, the amount of the loan, loan
Money purposes, the length of maturity.Each sample that the m cash borrows in sample include, but are not limited to age of creditor, educational background,
Annual income, the amount of the loan, the intended use of the loan, the length of maturity.
In the present embodiment, n and m are positive integer, and n is greater than m.
In one embodiment, institute can be obtained from the server (not shown) of communication connection by obtaining module 301
It states n normal credit requirement sample and the m cash borrows sample.The server is stored with the relevant data such as one of loan
The age of the credit information such as creditor for all previous loan that family or more banks are handled, educational background, annual income, the amount of the loan, loan
Money purposes, length of maturity etc..
Each sample in sample that execution module 302 borrows the sample of the n normal credit requirement and the m cash
Originally it is marked, comprising: be the n normal credit requirement sample using the n normal credit requirement sample as negative sample
In each sample increase a public field Sample_Flag, and Sample_Flag=-1 is set；The m cash is borrowed
Sample borrows each sample in sample for the m cash and increases a public field Sample_Flag as positive sample, and
Sample_Flag=1 is set.
The n normal credit requirement sample after label is borrowed sample with the m cash and mixed by execution module 302,
It is training set and verifying collection by mixed all random cuttings of sample, wherein it is first pre- to account for all samples for the training set
If ratio (such as 70%), the verifying collection accounts for the second preset ratio (such as 30%) of all samples.
Execution module 302 is using naive Bayesian (NB) as base classifier, with each sample included by the training set
The value (i.e. 1 and -1) of this public field Sample_Flag is as the tag along sort training base classifier.
Execution module 302 utilizes the trained base classifier, predicts the test set, exports the survey
It is the probability (that is to say the probability for exporting the Sample_Flag=1 of each sample) of positive sample that each sample is concentrated in examination.Execute mould
Block 302 carries out all samples included by the test set according to the size for the probability that each sample exported is positive sample
Descending arrangement, the preceding k normal credit requirement sample after filtering out descending arrangement, the k normal credit requirement that will be filtered out
Sample is as sample set D 's.The sample borrowed with the m cash has been filtered out from the n normal credit requirement sample as a result,
This most similar k normal credit requirement sample.
In one embodiment, k is less than or equal to the positive integer of m.Preferably, k is equal to m.
It should be noted that following description can be by the sample set D ' for convenience of the clear simple description present inventions
The included k normal credit requirement sample and the sample set DtThe included m cash is borrowed sample and is referred to as
" k+m sample ".Execution module 302 is using the k+m sample as total sample.
Execution module 302 is that initial weight is respectively set in each sample in the k+m sample.302 basis of execution module
The weight of each sample in the k+m sample obtains the weight vectors of each sample.
In the present embodiment, the initial weight that each sample is arranged in execution module 302 is 1.That is w1=w2=wk+m=1.The k
The weight vectors of+m samples
It is n times that maximum number of iterations, which is arranged, in execution module 302, and t indicates the number of iterations.Therefore, [1, N] t ∈.It changes when for the first time
The value of Dai Shi, t are equal to 1.
The weight of weight and each sample of the execution module 302 based on each sample in the k+m sample
Vector calculates weight ratio of each sample in the t times iteration.
In the present embodiment, weight ratio of each sample in the t times iteration is by formula: It is calculated.Wherein,Represent weight of each sample in the t times iteration in the k+m sample
Ratio,Represent weight vectors of each sample in the t times iteration in the k+m sample.Represent the k+m sample
Weight of each sample in the t times iteration in this.
For in other words, weight ratio of each sample in the t times iteration is equal to the k+m in the k+m sample
In a sample weight vectors of each sample in the t times iteration divided by the k+m sample the sum of weight.
Execution module 302 is by the sample set D 'sAs source sample, and by the sample set DtIt, will as target sample
The source sample D 'sWith target sample DtMixing (that is to say the k normal credit requirement sample and the m cash borrowing sample
This mixing).Execution module 302 is using linear regression (LR) as base classifier, by the mixed k+m sample in conjunction with described
Weight ratio of each sample in the t times iteration is trained the base classifier and obtains one with training in k+m sample
A Weak Classifier.
In the present embodiment, execution module 302 is arranged tag along sort 1 and represents high default risk, 0 represent in default risk, -1
Low default risk (i.e. different classification values represents different default risk degree) is represented, and remembers the Weak Classifier that the training obtains
Are as follows: ht: X → Y, Y ∈ { -1,0,1 }.
In the present embodiment, Weak Classifier htPass through each sample classification in the softmax layers of output k+m sample
To the Probability p of each tag along sort (i.e. 1,0, and -1), 0≤p≤1.
Execution module 302 calculates target sample D described in the Weak Classifier misclassificationtError rate.
Wherein, error rateWherein, | ht(xi)-c(xi) | indicate described weak point
Class device exports the target sample DtIn each sample be true tag along sort value probability and 1 difference.In other words ht(xi)
It represents the Weak Classifier and exports the target sample D in the t times iterationtIn each sample be true tag along sort value it is general
Rate, c (xi) value be considered as 1.It is describedRepresent the target sample DtIn weight of each sample in the t times iteration.
Execution module 302 is according to the target sample DtThe target sample D is calculated by the error rate of misclassificationtWeight
Renewal rate.Execution module 302 also calculates the source sample D 'sWeight renewal rate.
In the present embodiment, the target sample DtWeight renewal rateThe source sample D 'sWeight
Renewal rateWherein, n=k (that is to say the source sample D 'sIncluded letter by ordinary
With the quantity of loan sample), N is the set maximum number of iterations.
Execution module 302 is according to the target sample DtWeight renewal rate and the source sample D 'sWeight update
Rate determines the source sample D 'sWith target sample DtIn weight of each sample in the t+1 times iteration.
Specifically, weight of each sample in the t+1 times iteration are as follows:
Wherein,Weight of each sample in the t times iteration is represented,Each sample is then represented at the t+1 times
Weight when iteration.|ht(xi)-c(xi) | indicate that the Weak Classifier exports the target sample DtIn each sample be true
The probability of tag along sort value and 1 difference.
It can be seen that for the target sample DtIn sample improve the target sample D if classification errort
In weighted value of the sample in the t+1 times iteration, so that Weak Classifier is paid close attention to the sample for being classified mistake.And for described
Source sample D 'sIn sample reduce the source sample D ' if classification errorsIn power of the sample in the t+1 times iteration
Weight values, this is because classification error is considered as this partial data and target sample differs greatly.
Execution module 302 judges that currently whether having been carried out n times iteration (that is to say the value of the current t of comparison and the value of N
Size relation).If currently having not carried out n times iteration (i.e. current t < N), after t=t+1 is arranged in execution module 302, based on described
K+m sample described in the weight of each sample in k+m sample and the weight vector computation of the k+m sample is in t+
Weight ratio when 1 iteration.
If currently having been carried out n times iteration (i.e. current t=N), execution module 302 arrives iv-th iteration according to the N/2 times
The Weak Classifier obtained is trained (to that is to say that prediction is each to the prediction of the risk class of the k+m sample respectively in the process
The probability to each tag along sort of sample classification), it combines this N/2 times and trains weak point obtained in the process to iv-th iteration
Class device obtains a strong classifier.
For example, the strong classifier
The first condition refers to: the Weak Classifier that training obtains during described the N/2 times to iv-th iteration predicts institute
It states k+m sample classification and is greater than to the sum of the probability of tag along sort 1 and predict the k+m sample classification to the general of tag along sort 0
The sum of rate and it is greater than the sum of the probability value for predicting the k+m sample classification to tag along sort -1.The second condition refers to: institute
State the N/2 times predicts the k+m sample classification to tag along sort 0 to the Weak Classifier that training during iv-th iteration obtains
The sum of probability be greater than and predict described k+m sample classification to the sum of the probability of tag along sort 1 and greater than the prediction k+m sample
Originally it is categorized into the sum of the probability value of tag along sort -1.The third condition refers to: during described the N/2 times to iv-th iteration
The Weak Classifier that training obtains predicts the k+m sample classification to the sum of the probability of tag along sort -1 k+m described greater than prediction
A sample classification is to the sum of the probability of tag along sort 1 and is greater than the probability value for predicting the k+m sample classification to tag along sort 0
The sum of.
Execution module 302 using the strong classifier as risk identification model, using the risk identification model according to
The information of the loan transaction of audit identifies the risk class of pending loan transaction, and carries out air control according to recognition result.
In the present embodiment, the pending loan transaction is that cash borrows business.
The information of the pending loan transaction includes, but are not limited to, the age of creditor, educational background, annual income, loan
The amount of money, the intended use of the loan, the length of maturity.
For example, when predicting that a certain pending loan transaction is categorized into the probability of high default risk using the strong classifier
When greater than the probability for being categorized into middle default risk and being greater than the probability for being categorized into low default risk, execution module 302 determines this certain
Item loan transaction is high default risk business, and issues and be prompted to business personnel.
In one embodiment, when determining a certain loan transaction is high default risk business, execution module 302 may be used also
To determine the key message for influencing the risk class of a certain loan transaction using the risk identification model.Execution module 302
The key message is also prompted to business personnel, so that business personnel can be using these key messages as communicating with client
Reference information.
In one embodiment, described that the levels of risk for influencing a certain loan transaction is determined using the risk identification model
Other key message includes step (a1)-(a3):
(a1) information of a certain loan transaction is modified.
It for example, can be in the premise for keeping creditor's age, educational background, annual income, the intended use of the loan, the loan time limit constant
The lower modification amount of the loan such as reduces the amount of the loan.
(a2) it is identified using the risk identification model according to the information of a certain loan transaction described after modification modified
The risk class of a certain loan transaction.
(a3) when the risk class of the modified a certain loan transaction is lower than a certain loan transaction before modifying
When risk class, using the information modified as the key message for the risk class for influencing a certain loan transaction.In conclusion
Air control system described in the embodiment of the present invention based on transfer learning is borrowed first with SPY algorithm from a large amount of common credit
K sample similar with m target sample is filtered out in money sample, is then obtained by TrAdaBoost training for cash
The risk identification model for carrying out risk identification is borrowed, business can be borrowed to cash and carry out effective risk identification and control.
As shown in fig.3, the structural schematic diagram of the computer installation provided for present pre-ferred embodiments.The present invention compared with
In good embodiment, the computer installation 3 includes memory 31, at least one processor 32, at least one communication bus 33.This
Field technical staff both may be used it should be appreciated that the structure of the computer installation shown in Fig. 3 does not constitute the restriction of the embodiment of the present invention
To be bus topology, be also possible to star structure, the computer installation 3 can also include than illustrate it is more or fewer its
His hardware perhaps software or different component layouts.
In some embodiments, the computer installation 3 includes that one kind can be according to the instruction for being previously set or storing, certainly
The dynamic terminal for carrying out numerical value calculating and/or information processing, hardware include but is not limited to microprocessor, specific integrated circuit, can
Program gate array, digital processing unit and embedded device etc..
It should be noted that the computer installation 3 is only for example, other electronics that are existing or being likely to occur from now on are produced
Product are such as adaptable to the present invention, should also be included within protection scope of the present invention, and are incorporated herein by reference.
In some embodiments, the memory 31 is used to store program code and various data, such as is mounted on described
The air control system 30 based on transfer learning in computer installation 3, and in the operational process of computer installation 3 realize high speed,
It is automatically completed the access of program or data.The memory 31 include read-only memory (Read-Only Memory, ROM),
Random access memory (Random Access Memory, RAM), programmable read only memory (Programmable Read-Only
Memory, PROM), Erasable Programmable Read Only Memory EPROM (Erasable Programmable Read-Only Memory,
EPROM), disposable programmable read-only memory (One-time Programmable Read-Only Memory, OTPROM), electricity
Sub- erasing type can make carbon copies read-only memory (Electrically-Erasable Programmable Read-Only Memory,
EEPROM), CD-ROM (Compact Disc Read-Only Memory, CD-ROM) or other disc memories, disk are deposited
Reservoir, magnetic tape storage or any other the computer-readable medium that can be used in carrying or storing data.
In some embodiments, at least one described processor 32 can be made of integrated circuit, such as can be by single
The integrated circuit of encapsulation is formed, and is also possible to be made of the integrated circuit that multiple identical functions or different function encapsulate, be wrapped
Include one or more central processing unit (Central Processing unit, CPU), microprocessor, digital processing chip,
The combination etc. of graphics processor and various control chips.At least one described processor 32 is the control of the computer installation 3
Core (Control Unit), using all parts of various interfaces and the entire computer installation 3 of connection, by operation or
The program or module being stored in the memory 31 are executed, and calls the data being stored in the memory 31, with
The various functions and processing data of computer installation 3 are executed, such as execute the wind control function based on transfer learning.
In some embodiments, at least one described communication bus 33 is arranged to realize the memory 31 and described
Connection communication between at least one processor 32 etc..
Although being not shown, the computer installation 3 can also include the power supply (such as battery) powered to all parts, excellent
Choosing, power supply can be logically contiguous by electric power controller and at least one described processor 32, to pass through power management
Device realizes the functions such as management charging, electric discharge and power managed.Power supply can also include one or more direct current or
AC power source, recharging device, power failure detection circuit, power adapter or inverter, power supply status indicator etc. are appointed
Meaning component.The computer installation 3 can also include multiple sensors, bluetooth module, Wi-Fi module etc., and details are not described herein.
It should be appreciated that the embodiment is only purposes of discussion, do not limited by this structure in patent claim.
The above-mentioned integrated unit realized in the form of software function module, can store and computer-readable deposit at one
In storage media.Above-mentioned software function module is stored in a storage medium, including some instructions (i.e. one or more instructions)
With so that a computer installation (can be server, PC etc.) or processor (processor) execute the present invention
The part of each embodiment the method.
In a further embodiment, in conjunction with Fig. 2, the computer installation 3 is can be performed at least one described processor 32
Operating device and the types of applications program (the air control system 30 based on transfer learning as mentioned) of installation, program code
Deng for example, above-mentioned modules.
Program code is stored in the memory 31, and at least one described processor 32 can call the memory 31
The program code of middle storage is to execute relevant function.For example, modules described in Fig. 2 are stored in the memory 31
In program code, and as performed by least one described processor 32, to realize the function of the modules to reach
The purpose of air control based on transfer learning.
In one embodiment of the invention, the memory 31 stores one or more instructions, one or more of
Instruct the purpose that the air control based on transfer learning is realized performed by least one described processor 32.
Specifically, at least one described processor 32 can refer to figure to the concrete methods of realizing of said one or multiple instruction
The description of correlation step in 1 corresponding embodiment, this will not be repeated here.
In several embodiments provided by the present invention, it should be understood that disclosed device and method can pass through it
Its mode is realized.For example, the apparatus embodiments described above are merely exemplary, for example, the division of the module, only
Only a kind of logical function partition, there may be another division manner in actual implementation.
The module as illustrated by the separation member may or may not be physically separated, aobvious as module
The component shown may or may not be physical unit, it can and it is in one place, or may be distributed over multiple
In network unit.Some or all of the modules therein can be selected to realize the mesh of this embodiment scheme according to the actual needs
's.
It, can also be in addition, each functional module in each embodiment of the present invention can integrate in one processing unit
It is that each unit physically exists alone, can also be integrated in one unit with two or more units.Above-mentioned integrated list
Member both can take the form of hardware realization, can also realize in the form of hardware adds software function module.
It is obvious to a person skilled in the art that invention is not limited to the details of the above exemplary embodiments, Er Qie
In the case where without departing substantially from spirit or essential attributes of the invention, the present invention can be realized in other specific forms.Therefore, no matter
From the point of view of which point, the present embodiments are to be considered as illustrative and not restrictive, and the scope of the present invention is by appended power
Benefit requires rather than above description limits, it is intended that all by what is fallen within the meaning and scope of the equivalent elements of the claims
Variation is included in the present invention.Any reference signs in the claims should not be construed as limiting the involved claims.This
Outside, it is clear that one word of " comprising " is not excluded for other units or, odd number is not excluded for plural number.The multiple units stated in device claim
Or device can also be implemented through software or hardware by a unit or device.The first, the second equal words are used to indicate name
Claim, and does not indicate any particular order.
Finally it should be noted that the above examples are only used to illustrate the technical scheme of the present invention and are not limiting, although reference
Preferred embodiment describes the invention in detail, those skilled in the art should understand that, it can be to of the invention
Technical solution is modified or equivalent replacement, without departing from the spirit and scope of the technical solution of the present invention.
Claims (10)
1. a kind of air control method based on transfer learning, which is characterized in that the described method includes:
Obtaining step obtains source sample D 'sWith target sample Dt, wherein the source sample D 'sIncluding k sample, the target
Sample DtIncluding m sample, by the source sample D 'sWith target sample DtAs total sample, so that the number of total sample is equal to institute
State the sum of the number k of source sample and the number m of the target sample；
Initial weight is respectively set for each sample in total sample in first setting steps, and according in total sample
The weight of each sample obtain the weight vectors of each sample；
Second setting steps, setting maximum number of iterations are n times, and t indicates the number of iterations；
First executes step, based on the weight vectors of the weight of each sample in total sample and each sample
Calculate weight ratio of each sample in the t times iteration；
Second executes step, by the source sample D 'sWith the target sample DtMixing obtains mixed total sample, with linear
Return and be used as base classifier, by mixed total sample in conjunction with weight ratio of each sample in the t time iteration it is trained described in
Base classifier obtains a Weak Classifier；
Third executes step, calculates target sample D described in the Weak Classifier misclassificationtError rate；
4th executes step, calculates the source sample D 'sWeight renewal rate, and according to the target sample DtAccidentally divided
The error rate of class calculates the target sample DtWeight renewal rate；
5th executes step, according to the target sample DtWeight renewal rate and the source sample D 'sWeight update speed
Rate determines the source sample D 'sWith target sample DtIn weight of each sample in the t+1 times iteration；
Whether judgment step, judgement currently have been carried out n times iteration, wherein if currently having not carried out n times iteration, t is arranged
Described first is triggered after=t+1 executes step；If currently having been carried out n times iteration, triggering the 6th executes step；
Described 6th executes step, trains the Weak Classifier obtained respectively to institute in the process according to the N/2 times to iv-th iteration
The prediction of the risk class of total sample combines this N/2 times and trains the Weak Classifier obtained to obtain in the process to iv-th iteration
One strong classifier；And
7th executes step, using the strong classifier as risk identification model, using the risk identification model according to pending
The information of the loan transaction of core identifies the risk class of pending loan transaction, and carries out air control according to recognition result.
2. the air control method based on transfer learning as described in claim 1, which is characterized in that the target sample DtIt is included
The m sample be cash borrow sample, the source sample D 'sThe included k sample is normal credit requirement sample,
This method obtains the source sample D 'sThe step of include:
Obtain sample set Ds, wherein the sample set DsSample including n normal credit requirement；
Each sample in sample that sample and the m cash to the n normal credit requirement are borrowed is marked, and wraps
It includes: being each sample in the n normal credit requirement sample using the n normal credit requirement sample as negative sample
Increase a public field Sample_Flag, and Sample_Flag=-1 is set；The m cash is borrowed into sample as positive sample
This, borrows each sample in sample for the m cash and increases a public field Sample_Flag, and Sample_ is arranged
Flag=1；
The n normal credit requirement sample after label is borrowed sample with the m cash to mix, by mixed all samples
This random cutting is training set and verifying collection；
Using naive Bayesian as base classifier, with the public field Sample_ of each sample included by the training set
The value of Flag is as the tag along sort training base classifier；
The test set is predicted using the trained base classifier, exporting each sample in the test set is
The probability of positive sample；
All samples included by the test set are carried out according to the size for the probability that each sample exported is positive sample
Descending arrangement, the preceding k normal credit requirement sample after filtering out descending arrangement, the k normal credit requirement that will be filtered out
Sample is as the source sample D 's。
3. the air control method based on transfer learning as described in claim 1, which is characterized in that each sample in total sample
Originally the weight ratio in the t times iteration is by formula:It is calculated, whereinRepresent total sample
In weight ratio of each sample in the t times iteration,Each sample in total sample is represented in the t times iteration
Weight vectors,Represent weight of each sample in the t times iteration in total sample.
4. the air control method based on transfer learning as described in claim 1, which is characterized in that described second executes in step:
Setting tag along sort 1 represents high default risk, and default risk in 0 representative, -1 represents low default risk, and remembers what the training obtained
Weak Classifier are as follows: ht: X → Y, Y ∈ { -1,0,1 }；Wherein, Weak Classifier htPass through the softmax layers of output k+m sample
In Probability p of each sample classification to each tag along sort, 0≤p≤1.
5. the air control method based on transfer learning as described in claim 1, which is characterized in that the error rate is according to formulaIt is calculated, wherein ∈tThe error rate is represented, | ht(xi)-c(xi) | indicate institute
It states Weak Classifier and exports the target sample DtIn each sample be true tag along sort value probability and 1 difference, it is described
Represent the target sample DtIn weight of each sample in the t times iteration.
6. the air control method based on transfer learning as described in claim 1, which is characterized in that the target sample DtWeight
Renewal rateThe source sample D 'sWeight renewal rate Wherein, n=
K, N are the set maximum number of iterations.
7. the air control method based on transfer learning as described in claim 1, which is characterized in that each sample changes at the t+1 times
For when weight are as follows:
Wherein,Weight of each sample in the t times iteration is represented,Each sample is then represented in the t+1 times iteration
Weight, | ht(xi)-c(xi) | indicate that the Weak Classifier exports the target sample DtIn each sample be true contingency table
The probability of label value and 1 difference.
8. a kind of computer installation, which is characterized in that the computer installation includes memory and processor, and the memory is used
In storing at least one instruction, the processor when at least one described instruction for realizing as any one in claim 1 to 7
Air control method based on transfer learning described in.
9. a kind of computer readable storage medium, which is characterized in that the computer-readable recording medium storage has at least one
Instruction, at least one described instruction are realized when being executed by processor and are learned as described in any one of claim 1 to 7 based on migration
The air control method of habit.
10. a kind of air control system based on transfer learning, runs on computer installation, which is characterized in that the system comprises:
Module is obtained, for obtaining source sample D 'sWith target sample Dt, wherein the source sample D 'sIt is described including k sample
Target sample DtIncluding m sample, by the source sample D 'sWith target sample DtAs total sample, thus the number etc. of total sample
In the sum of the number k and the number m of the target sample of the source sample；
Execution module, for initial weight to be respectively set for each sample in total sample, and according in total sample
The weight of each sample obtain the weight vectors of each sample；
The execution module, being also used to be arranged maximum number of iterations is n times, and t indicates the number of iterations；
The execution module is also used to the weight of weight and each sample based on each sample in total sample
Vector calculates weight ratio of each sample in the t times iteration；
The execution module is also used to the source sample D 'sWith the target sample DtMixing obtains mixed total sample,
Weight ratio using linear regression as base classifier, by mixed total sample in conjunction with each sample in the t times iteration
The rate training base classifier obtains a Weak Classifier；
The execution module is also used to calculate target sample D described in the Weak Classifier misclassificationtError rate；
The execution module is also used to calculate the source sample D 'sWeight renewal rate, and according to the target sample Dt
The target sample D is calculated by the error rate of misclassificationtWeight renewal rate；
The execution module is also used to according to the target sample DtWeight renewal rate and the source sample D ' s weight
Renewal rate determines the source sample D 'sWith target sample DtIn weight of each sample in the t+1 times iteration；
The execution module is also used to currently whether have been carried out n times iteration, wherein if currently having not carried out n times iteration,
The first execution step is triggered after t=t+1 is then set；If currently have been carried out n times iteration, triggering the 6th executes step
Suddenly；
The execution module is also used to train the Weak Classifier obtained right respectively in the process according to the N/2 times to iv-th iteration
The prediction of the risk class of total sample combines this N/2 times and trains the Weak Classifier obtained in the process to iv-th iteration
Obtain a strong classifier；And
The execution module is also used to utilize the risk identification model root using the strong classifier as risk identification model
The risk class of pending loan transaction is identified according to the information of pending loan transaction, and wind is carried out according to recognition result
Control.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910498007.3A CN110223164A (en) | 2019-06-10 | 2019-06-10 | Air control method and system based on transfer learning, computer installation, storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910498007.3A CN110223164A (en) | 2019-06-10 | 2019-06-10 | Air control method and system based on transfer learning, computer installation, storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110223164A true CN110223164A (en) | 2019-09-10 |
Family
ID=67816106
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910498007.3A Pending CN110223164A (en) | 2019-06-10 | 2019-06-10 | Air control method and system based on transfer learning, computer installation, storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110223164A (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111080123A (en) * | 2019-12-14 | 2020-04-28 | 支付宝(杭州)信息技术有限公司 | User risk assessment method and device, electronic equipment and storage medium |
| CN111429258A (en) * | 2020-03-19 | 2020-07-17 | 中国建设银行股份有限公司 | Method and device for monitoring loan fund flow direction |
| CN111429257A (en) * | 2020-03-19 | 2020-07-17 | 中国建设银行股份有限公司 | Transaction monitoring method and device |
| CN112750038A (en) * | 2021-01-14 | 2021-05-04 | 中国工商银行股份有限公司 | Transaction risk determination method and device and server |
| WO2021169115A1 (en) * | 2020-02-29 | 2021-09-02 | 平安科技（深圳）有限公司 | Risk control method, apparatus, electronic device, and computer-readable storage medium |
| CN113610176A (en) * | 2021-08-16 | 2021-11-05 | 上海冰鉴信息科技有限公司 | Cross-scene migration classification model forming method and device and readable storage medium |
-
2019
- 2019-06-10 CN CN201910498007.3A patent/CN110223164A/en active Pending
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111080123A (en) * | 2019-12-14 | 2020-04-28 | 支付宝(杭州)信息技术有限公司 | User risk assessment method and device, electronic equipment and storage medium |
| WO2021169115A1 (en) * | 2020-02-29 | 2021-09-02 | 平安科技（深圳）有限公司 | Risk control method, apparatus, electronic device, and computer-readable storage medium |
| CN111429258A (en) * | 2020-03-19 | 2020-07-17 | 中国建设银行股份有限公司 | Method and device for monitoring loan fund flow direction |
| CN111429257A (en) * | 2020-03-19 | 2020-07-17 | 中国建设银行股份有限公司 | Transaction monitoring method and device |
| CN111429257B (en) * | 2020-03-19 | 2024-04-12 | 中国建设银行股份有限公司 | Transaction monitoring method and device |
| CN112750038A (en) * | 2021-01-14 | 2021-05-04 | 中国工商银行股份有限公司 | Transaction risk determination method and device and server |
| CN112750038B (en) * | 2021-01-14 | 2024-02-02 | 中国工商银行股份有限公司 | Transaction risk determination method, device and server |
| CN113610176A (en) * | 2021-08-16 | 2021-11-05 | 上海冰鉴信息科技有限公司 | Cross-scene migration classification model forming method and device and readable storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110223164A (en) | Air control method and system based on transfer learning, computer installation, storage medium | |
| CN108764984A (en) | A kind of power consumer portrait construction method and system based on big data | |
| Cheng et al. | Towards better forecasting by fusing near and distant future visions | |
| CN107818344A (en) | The method and system that user behavior is classified and predicted | |
| US20210133490A1 (en) | System and method for unsupervised abstraction of sensitive data for detection model sharing across entities | |
| CN109325640A (en) | User's Value Prediction Methods, device, storage medium and equipment | |
| CN101546290B (en) | Method for improving accuracy of quality forecast of class hierarchy in object-oriented software | |
| CN115330275B (en) | Echelon utilization method and device for retired battery | |
| CN108711047A (en) | A kind of automatic repayment method, system and terminal device | |
| CN110264270A (en) | A kind of behavior prediction method, apparatus, equipment and storage medium | |
| CN109829721A (en) | Online trading multiagent behavior modeling method based on heterogeneous network representative learning | |
| García-Vico et al. | Fepds: A proposal for the extraction of fuzzy emerging patterns in data streams | |
| CN113139743A (en) | Sewage discharge index analysis method and device, electronic equipment and storage medium | |
| CN113626606A (en) | Information classification method and device, electronic equipment and readable storage medium | |
| CN109345371A (en) | Personal reference report backtracking method and system | |
| US11854018B2 (en) | Labeling optimization through image clustering | |
| CN110163517A (en) | It is a kind of to be had a holiday regular block catenary system modeling method based on G- limitation service | |
| CN112766814A (en) | Training method, device and equipment for credit risk pressure test model | |
| Zheng et al. | Fairness-aware multi-view clustering | |
| US20210133586A1 (en) | System and method for unsupervised abstraction of sensitive data for realistic modeling | |
| US20210133488A1 (en) | System and method for unsupervised abstraction of sensitive data for consortium sharing | |
| US20210133783A1 (en) | System and method for unsupervised abstraction of sensitive data for detection model sharing across entities | |
| CN115168848A (en) | Interception feedback processing method based on big data analysis interception | |
| Le Lay et al. | Automated generation of patient population for discrete-event simulation using process mining | |
| CN114662794A (en) | Enterprise default risk prediction method, device, equipment and storage medium |
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: 20190910 |