CN108573358A - A kind of overdue prediction model generation method and terminal device - Google Patents
A kind of overdue prediction model generation method and terminal device Download PDFInfo
- Publication number
- CN108573358A CN108573358A CN201810436131.2A CN201810436131A CN108573358A CN 108573358 A CN108573358 A CN 108573358A CN 201810436131 A CN201810436131 A CN 201810436131A CN 108573358 A CN108573358 A CN 108573358A
- Authority
- CN
- China
- Prior art keywords
- user
- overdue
- prediction model
- random
- model
- 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
- 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
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0635—Risk analysis of enterprise or organisation activities
-
- 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
- Business, Economics & Management (AREA)
- Engineering & Computer Science (AREA)
- Human Resources & Organizations (AREA)
- Economics (AREA)
- Strategic Management (AREA)
- General Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- Development Economics (AREA)
- Entrepreneurship & Innovation (AREA)
- Physics & Mathematics (AREA)
- Finance (AREA)
- General Physics & Mathematics (AREA)
- Marketing (AREA)
- Accounting & Taxation (AREA)
- Educational Administration (AREA)
- Technology Law (AREA)
- Game Theory and Decision Science (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The present invention provides a kind of overdue prediction model generation method and terminal devices, are suitable for technical field of data processing, and this method includes：User attribute data is screened by learning model, is obtained and the maximum N kinds user attribute data of the overdue class label degree of association；Random grouping is carried out to user and obtains the random training sample matrix of multiple numbers of users, then is trained to obtain corresponding multiple sub- prediction models respectively；Based on comprising number of users calculate the ballot coefficient of sub- prediction model, and build and obtain required overdue prediction model.The embodiment of the present invention realizes the prediction of the efficiently and accurately overdue to user.
Description
Technical field
The invention belongs to technical field of data processing more particularly to overdue prediction model generation method and terminal devices.
Background technology
Risk management after providing a loan in the prior art user is relatively simple, is all that passively whether acquisition user is overdue
Not also as a result, and carried out again after determining that user is overdue subsequent collection or Claims Resolution processing.However when user exceedes
When the case where phase does not go back, borrower's company has had been subjected to certain economic loss, while carrying out promise breaking fund to overdue user
Recovery difficult is larger so that borrower's company's economic loss further aggravates.
To sum up, it cannot achieve in the prior art and effective assessment prediction carried out to the overdue risk of user.
Invention content
In view of this, an embodiment of the present invention provides a kind of overdue prediction model generation method and terminal device, to solve
The problem of effective assessment prediction can not being carried out to the overdue risk of user in the prior art.
The first aspect of the embodiment of the present invention provides a kind of overdue prediction model generation method, including：
The L kinds user attribute data of M user of acquisition and overdue class label obtain the input that dimension is M × (L+1)
Matrix, wherein M and L is positive integer；
The input matrix is input in preset learning model, is filtered out with the overdue class label degree of association most
Big N kind user attribute datas, the learning model are the study for carrying out user property and overdue class label correlation analysis
Model, wherein N is the positive integer less than or equal to L；
The M user is grouped at random, obtains number of users PhH random user group, and generate the H use
Group corresponding dimension in family is PhThe H training sample matrix of × (N+1), wherein PhFor positive integer, H is just whole more than 1
Number, h ∈ [1, H]；
The H training sample matrix is inputted respectively in preset H neural network model and is trained until meeting
The preset condition of convergence determines the corresponding model parameter of the H neural network model, obtains H sub- prediction models；
The number of users P for including according to the H user grouphThe H corresponding H of sub- prediction models are calculated to throw
Ticket coefficient, and obtain overdue prediction based on the H sub- prediction models and the corresponding H ballot coefficient, structure
Model：
Presult=Vcoent1Vresult1+Vcoent2Vresult2+…
+VcoenthVresulth+…+VcoentHVresultH
Wherein, Presult is the output of the overdue prediction model as a result, VresulthFor h-th of sub- prediction mould
The output of type is as a result, VcoenthFor the corresponding ballot coefficient of h-th of sub- prediction model.
The second aspect of the embodiment of the present invention provides a kind of overdue prediction model generation terminal device, the overdue prediction
It includes memory, processor that model, which generates terminal device, and the meter that can be run on the processor is stored on the memory
Calculation machine program, the processor realize following steps when executing the computer program.
The L kinds user attribute data of M user of acquisition and overdue class label obtain the input that dimension is M × (L+1)
Matrix, wherein M and L is positive integer；
The input matrix is input in preset learning model, is filtered out with the overdue class label degree of association most
Big N kind user attribute datas, the learning model are the study for carrying out user property and overdue class label correlation analysis
Model, wherein N is the positive integer less than or equal to L；
The M user is grouped at random, obtains number of users PhH random user group, and generate the H use
Group corresponding dimension in family is PhThe H training sample matrix of × (N+1), wherein PhFor positive integer, H is just whole more than 1
Number, h ∈ [1, H]；
The H training sample matrix is inputted respectively in preset H neural network model and is trained until meeting
The preset condition of convergence determines the corresponding model parameter of the H neural network model, obtains H sub- prediction models；
The number of users P for including according to the H user grouphThe H corresponding H of sub- prediction models are calculated to throw
Ticket coefficient, and obtain overdue prediction based on the H sub- prediction models and the corresponding H ballot coefficient, structure
Model：
Presult=Vcoent1Vresult1+Vcoent2Vresult2+…
+VcoenthVresulth+…+VcoentHVresultH
Wherein, Presult is the output of the overdue prediction model as a result, VresulthFor h-th of sub- prediction mould
The output of type is as a result, VcoenthFor the corresponding ballot coefficient of h-th of sub- prediction model.
The third aspect of the embodiment of the present invention provides a kind of computer readable storage medium, including：It is stored with computer
Program, which is characterized in that overdue prediction model generation side as described above is realized when the computer program is executed by processor
The step of method.
Existing advantageous effect is the embodiment of the present invention compared with prior art：By to the adaptive of user attribute data
Study, identify wherein with whether the once overdue maximum N kinds user attribute data of the overdue class label degree of association, realize
To most preferably be used for overdue prediction user property identification, ensure that subsequently to user whether can overdue prediction efficient process.
By the way that user be grouped at random and trained, to obtain the submodel that can be used for the overdue prediction of user of random amount, most
Calculate the corresponding ballot coefficient of each overdue prediction submodel afterwards, and by all overdue prediction submodels and corresponding throwing
Ticket coefficient package is overdue prediction model so that overdue prediction model, can be by each when carrying out overdue prediction to user
The prediction result of submodel is voted, and the weight calculation of final vote result is carried out via corresponding ballot coefficient, to
It ensure that the accurate and reliable of final prediction result.Therefore, the embodiment of the present invention realize whether can be overdue to user efficient standard
True prediction.
Description of the drawings
It to describe the technical solutions in the embodiments of the present invention more clearly, below will be to embodiment or description of the prior art
Needed in attached drawing be briefly described, it should be apparent that, the accompanying drawings in the following description be only the present invention some
Embodiment for those of ordinary skill in the art without having to pay creative labor, can also be according to these
Attached drawing obtains other attached drawings.
Fig. 1 is the implementation process schematic diagram for the overdue prediction model generation method that the embodiment of the present invention one provides；
Fig. 2 is the implementation process schematic diagram of overdue prediction model generation method provided by Embodiment 2 of the present invention；
Fig. 3 is the implementation process schematic diagram for the overdue prediction model generation method that the embodiment of the present invention four provides；
Fig. 4 is the implementation process schematic diagram for the overdue prediction model generation method that the embodiment of the present invention five provides；
Fig. 5 is the structural schematic diagram for the overdue prediction model generating means that the embodiment of the present invention six provides；
Fig. 6 is the schematic diagram that the overdue prediction model that the embodiment of the present invention seven provides generates terminal device.
Specific implementation mode
In being described below, for illustration and not for limitation, it is proposed that such as tool of particular system structure, technology etc
Body details, to understand thoroughly the embodiment of the present invention.However, it will be clear to one skilled in the art that there is no these specific
The present invention can also be realized in the other embodiments of details.In other situations, it omits to well-known system, device, electricity
The detailed description of road and method, in case unnecessary details interferes description of the invention.
In order to illustrate technical solutions according to the invention, illustrated below by specific embodiment.
Fig. 1 shows the implementation flow chart for the overdue prediction model generation method that the embodiment of the present invention one provides, and is described in detail such as
Under：
S101, the L kinds user attribute data of M user of acquisition and overdue class label, it is M × (L+1) to obtain dimension
Input matrix, wherein M and L is positive integer.
Wherein, user attribute data refers to age of user, gender, occupation, personal annual income, annual family income, loan
The userspersonal informations such as situation and occupancy, these user attribute datas can be manually entered by technical staff or from debt-credits
It chooses and obtains in the stored userspersonal information of company.Wherein, it is contemplated that require to carry when every borrower's company borrows money to user
The userspersonal information of confession may be different, and the userspersonal information of the audit logging needed for different loan products also may be used
Can difference, if small amount loan product is less with respect to the userspersonal information needed for wholesale loan product, therefore, the present invention is real
The type for applying quantity L and the userspersonal information that specifically includes of the example not to L kind user attribute datas is defined, can be by
Technical staff chooses according to actual demand.Overdue class label refers to the record that whether there is overdue behavior to user
Label, including there is no overdue behavior and can be used there are overdue behavior two categories, in practical application Chinese character/English/
The forms such as number record label, such as " 0 " are used to indicate there is no overdue behavior, and using " 1 " expression, there are overdue behaviors to save
Save memory.Wherein, due to the embodiment of the present invention both can be used for whether class loan product single to user/mono- can be overdue it is pre-
Survey, can be used for user's universal class loan product whether prediction that can be overdue, therefore the overdue classification in the embodiment of the present invention
Label, either user it is single/mono- class loan product whether there is the record label of overdue behavior, as user borrows small amount
The record label that product whether there is overdue behavior is borrowed, can also be that user's universal class loan product whether there is overdue behavior
Record label, if user whether there is the record label of overdue behavior, specifically need to by technical staff borrowing according to actual prediction
It is selected to borrow product.
In embodiments of the present invention, in order to ensure subsequently to prediction model training structure validity ensure prediction model pair
The accuracy of overdue prediction, number of users M do not answer very few, preferably may be configured as 100,000.
Input matrix is input in preset learning model by S102, is filtered out and overdue class label degree of association maximum
N kind user attribute datas, learning model be carry out user property and overdue class label correlation analysis learning model,
In, N is the positive integer less than or equal to L.
Since the userspersonal information's type got in actual conditions is generally all more, and and it is not all with user
Whether overdue association can be had, if therefore directly carrying out subsequent prediction model using all userspersonal informations got
Training structure, on the one hand can bring great live load to training structure, on the other hand also can strong influence it is final
The validity of the prediction model arrived reduces the accuracy rate of prediction model.Therefore, it is carried to promote the efficiency built to prediction model
For the accuracy rate of prediction model, the embodiment of the present invention can utilize adaptive learning technology to the userspersonal information that gets and exceed
Phase class label is learnt, and is extracted from L kind userspersonal informations and the highest N kinds user of the overdue behavior degree of association of user
Personal information, using as subsequent training data.Wherein, the learning model actually used is either some existing correlations
Analysis model can also be not to be limited herein by the learning model of technical staff's designed, designed, preferably can refer to the present invention
Embodiment two is handled, and the occurrence of N need to be by determining after learning model analyzing processing.
S103 is grouped M user at random, obtains number of users PhH random user group, and generate H user
The corresponding dimension of group is PhThe H training sample matrix of × (N+1), wherein PhFor positive integer, H is the positive integer more than 1,
H ∈ [1, H].
In the embodiment of the present invention, can be used can put back to formula or can not put back to the methods of sampling of formula to realize that user divides at random
Group.Wherein, it is contemplated that if very few in the user's sample number used when prediction model training, the prediction that may result in
Model accuracy rate is relatively low, therefore, a smallest sample number can preferably be arranged in embodiments of the present invention, and carrying out user's pumping
Sample requires every group of number of users to be equal to smallest sample greatly when being grouped, with the number of training of each prediction model ensured
It will not be very few.
H training sample matrix is inputted in preset H neural network model and is trained until meeting by S104 respectively
The preset condition of convergence determines the corresponding model parameter of H neural network model, obtains H sub- prediction models.
After obtaining H training sample, the embodiment of the present invention can carry out model training respectively for each training sample,
Obtain corresponding H sub- prediction models.Wherein, the training method and the corresponding condition of convergence specifically used can be by technology people
Member is chosen and is set according to actual demand, including but not limited to such as using common weight adaptive learning algorithm come to instruction
Practice sample and carry out weight study processing, and set the condition of convergence to success rate prediction and be more than threshold value, then is adaptively learned in weight
The corresponding weighted value of each userspersonal information in training sample is determined after the completion of practising, and obtains corresponding sub- prediction mould
Type.Since all training structure obtains every sub- prediction model independently of each other in the embodiment of the present invention, each height is pre-
Survey model can be achieved to user whether prediction that can be overdue, and obtain corresponding prediction result.
S105, the number of users P for including according to H user grouphH corresponding H of sub- prediction models are calculated to vote
Coefficient, and obtain overdue prediction model based on H sub- prediction models and corresponding H ballot coefficient, structure.
Wherein, overdue prediction model formula is：
Presult=Vcoent1Vresult1+Vcoent2Vresult2+…
+VcoenthVresulth+…+VcoentHVresultH (1)
Presult is the output of overdue prediction model as a result, VresulthFor h-th of sub- prediction model output as a result,
VcoenthFor the corresponding ballot coefficient of h-th of sub- prediction model.
After obtaining H sub- prediction models, the embodiment of the present invention can be packaged structure based on this little prediction model and exceed
Phase prediction model, and the prediction knot of final overdue prediction model is determined by the way of the ballot of sub- prediction model prediction result
Fruit.
Due to number of training number with the obtained predictablity rate of prediction model have direct association, and instructing
The accuracy rate of its more obtained prediction model of number of training also will be higher when white silk method is identical, therefore, in order to be promoted most
The accuracy rate of the overdue prediction model obtained eventually, the embodiment of the present invention using sub- prediction model carry out result ballot prediction when,
The calculating of ballot coefficient can be carried out to determine that every height is predicted according to the number of users of the corresponding training sample of sub- prediction model
Proportion of the model result in ballot, then proportion conversion is carried out simultaneously to the result of every sub- prediction model based on these ballot coefficients
Final voting results are counted, final prediction result is obtained.Wherein, vote coefficient computational methods include but not limited to as will
The corresponding number of users P of sub- prediction modelhDivided by total number of users M, corresponding ballot coefficient is obtained, such as assumes total number of users M=
10000, the number of users P of first sub- prediction model1=5000, at this time the corresponding ballot coefficient of first sub- prediction model=
5000/10000=0.5.
It is illustrated with an example, it is assumed that there are the first sub- prediction model to the 4th sub- prediction model totally 4 sub- prediction models,
Corresponding ballot coefficient is respectively 0.5,0.3,0.4 and 0.7, and sub- prediction model prediction result " will produce overdue behavior " is remembered
It is 1, " not will produce overdue behavior " is denoted as the -1, and first sub- prediction model to the prediction result point of the 4th sub- prediction model output
Not Wei 1, -1,1 and -1, at this time according to formula (1) it is found that the final voting results=0.5-0.3+0.4- of overdue prediction model
0.7=-0.1<0, therefore, the final prediction result of overdue prediction model should be user and not will produce overdue behavior at this time.
The embodiment of the present invention by the adaptive learning to user attribute data, identify wherein with the presence or absence of overdue row
For the maximum N kinds user attribute data of the overdue class label degree of association, realize to most preferably be used for overdue prediction user belong to
The identification of property, ensure that subsequently to the efficient process of the overdue prediction of user, and ensure that the accuracy rate of prediction.By to user into
Row is random to be grouped and trains, and to obtain the submodel that can be used for the overdue prediction of user of random amount, is finally calculated and is each exceeded
Phase predicts the corresponding ballot coefficient of submodel, and is by all overdue prediction submodels and corresponding ballot coefficient package
Overdue prediction model so that overdue prediction model can pass through the prediction of each submodel when carrying out overdue prediction to user
As a result it votes, and carries out the weight calculation of final vote result via corresponding ballot coefficient, it is final pre- to ensure that
Survey the accurate and reliable of result.Therefore the embodiment of the present invention realize to user whether the prediction of efficiently and accurately that can be overdue.
As a kind of concrete methods of realizing of learning model, as shown in Fig. 2, the embodiment of the present invention two includes：
S201 calculates the corresponding comentropy of M user based on overdue class label.
Wherein, the calculation formula of the corresponding comentropy of M user is as follows：
MUoverTo there is no the number of users of overdue behavior, M in M userOverFor there are the use of overdue behavior in M user
Amount.
S202 carries out two dimensionization processing to L kind user attribute datas, and calculates the corresponding division of L kind user attribute datas
Comentropy and division information.
Two-dimentionalization processing refers to that user attribute data is divided into two class states, and specified wherein one according to certain rule
Class state is first kind state, and gender two dimension is such as divided into male, two class state of women, and it is first kind shape to specify male
State, has been divided into room or without room by occupancy, and debt does not pay off, has room and debt to pay off two class states, be assigned with room and
Debt is paid off as first kind state.At can only be to the data comprising two states when carrying out division information entropy and calculating
Reason, but the possible more than two state of user attribute data result in actual conditions, as the occupancy of user includes no room, has by oneself
House, debt has been paid off and self-owned house, and debt does not pay off three kinds of states, can not directly carry out the meter of division information entropy at this time
It calculates, therefore all user attribute data two dimensionizations can be handled in the embodiment of the present invention, to ensure the normal place of follow-up data
Reason.Wherein, the specified rule of the rule of division and first kind state, can be set according to actual conditions by technical staff.
The calculation formula of the division information entropy of L kind user attribute datas is as follows：
The calculation formula of the division information of L kind user attribute datas is as follows：
AttriRefer to the number of users of its first kind state after i-th kind of user attribute data two dimension, such as gender two dimension
Later in M users male quantity.Attri1It refer to first kind status user after i-th kind of user attribute data two dimension
In overdue number of users, the overdue number of users such as male in examples detailed above.
S203 calculates the corresponding information gain of M user based on comentropy and division information entropy.
The calculation formula of the corresponding information gain of M user is as follows：
Gain (M)=Entropy (M)-Entropy (M, Attr) (5)
S204 calculates the corresponding information gain-ratio of L kind user attribute datas based on division information and information gain.
The corresponding information gain-ratio calculation formula of L kind user attribute datas is as follows：
S205 randomly selects out I to L-1 kind user attribute datas from L kind user attribute datas, and is based on randomly selecting
The user attribute data arrived returns to the operation for executing and calculating the corresponding comentropy of M user based on overdue class label, to obtain
Corresponding multiple information gain-ratios, wherein I are the positive integer less than or equal to L-1.
S206 finds out the corresponding user attribute data of maximum information gain-ratio in multiple information gain-ratios, with determination
Go out N kind user attribute datas.
Since information gain-ratio is bigger, show that whether overdue its corresponding user attribute data and user's correlation be bigger,
Therefore, in order to determine with the maximum N kinds user attribute data of the overdue class label degree of association, theoretically need calculate L kinds use
The corresponding information gain-ratio of the various number combinations of family attribute data, and the wherein corresponding user attribute data combination of maximum value is taken
As required N kind user attribute datas.It is right however by practical application it is found that due to when user attribute data is considerably less
User whether prediction that can be overdue will be inaccurate, such as a kind of single user attribute data necessarily can not Accurate Prediction go out user
Whether can be overdue, therefore, in order to reduce unnecessary operand, a user attribute data number can be preset in the embodiment of the present invention
Minimum value I includes at least I kind user attribute datas for every group, you can obtain when carrying out random combine Group user attribute data, in addition scriptIt is available altogetherA information
Ratio of profit increase.
As the embodiment of the present invention three, when progress user is grouped at random, including：
If M is greater than or equal to number of users threshold value, M is used using that can put back to formula or the methods of sampling of formula can not be put back to
Family carries out random sampling, to realize the random grouping to M user, obtains comprising H random user group of number of users.
If M is less than number of users threshold value, random sampling is carried out to M user using the methods of sampling that can put back to formula, with reality
Now to the random grouping of M user, obtain comprising H random user group of number of users.
Wherein, number of users threshold value can be configured by technical staff according to actual conditions.When M is equal to number of users threshold value greatly
Illustrating that number of users is bigger, the methods of sampling that can either put back to formula still at this time and can not put back to formula carries out user and is grouped at random,
It is all easier to ensure that every group of number of users will not be very few, the accuracy rate of subsequent prediction model will not be impacted, but work as user
Number M is smaller, when being less than number of users threshold value, when being grouped, most probably divides according to the methods of sampling that can not put back to formula
The less situation of the number of users that arrives, at this time may have some impact on the accuracy rate of subsequent prediction model, therefore the present invention
It can be at random grouped only with the methods of sampling that can put back to formula when number of users is smaller in embodiment, to ensure prediction model
Accuracy rate.
As the embodiment of the present invention four, as shown in figure 3, after obtaining overdue prediction model, further include：
S301, obtains the predictablity rate of the overdue prediction model of output in preset time period, and judges that predictablity rate is
It is no to be less than accuracy rate threshold value.
When predictablity rate is relatively low, the overdue prediction model illustrated has been difficult to adapt to actual forecast demand,
Therefore it needs to carry out model modification to it, to ensure the real-time accurate and effective of overdue prediction model.
Wherein, predictablity rate can both be directly inputted by technical staff, can also be the prediction read in preset time period
It is obtained after overdue result and the overdue interpretation of result of corresponding reality.Preset time period and accuracy rate threshold value need to be by technical staff
It is configured according to actual demand situation, it is preferable that preset time period may be configured as nearest half a year.
S302 obtains the N kind user property numbers of T user if judging result, which is predictablity rate, is less than accuracy rate threshold value
According to this and overdue class label, the sample matrix that dimension is T × (N+1) is obtained, wherein T is positive integer.
Due to being had determined that in the embodiment of the present invention one and the highest N kinds user property number of the overdue class label degree of correlation
According to therefore, no longer needing to be analyzed using learning model at this time, it is only necessary to directly read the corresponding N kinds user of each user and belong to
Property data.
S303 is grouped T user at random, obtains number of users PhH random user group, and generate H user
The corresponding dimension of group is PhThe H training sample matrix of × (N+1), wherein PhFor positive integer, H is the positive integer more than 1,
H ∈ [1, H].
S304, return are executed to input H training sample matrix in preset H neural network model respectively and be trained
Until meeting the preset condition of convergence, the corresponding model parameter of H neural network model is determined, obtain H son prediction
The operation of model, to carry out model modification to overdue prediction model.
After the user attribute data for obtaining new user, the user attribute data of new user is recycled to come to overdue pre-
It surveys model and re-starts update, ensure the real-time accurate and effective of overdue prediction model.
As the embodiment of the present invention five, as shown in figure 4, the prediction for obtaining the overdue prediction model of output in preset time period is accurate
True rate, and judge whether predictablity rate is less than accuracy rate threshold value, including：
S401 parses loan product input by user and chooses the pointed loan product type of instruction.
S402 parses debt-credit from the predictablity rate for exporting overdue prediction model in the preset time period got
The corresponding predictablity rate of product type.
S403, judges whether the corresponding predictablity rate of loan product type is less than accuracy rate threshold value.
In view of in actual conditions, the embodiment of the present invention one to four may be the overdue prediction to multiclass loan product,
The overdue prediction result of user of many different loan products may be included in obtained prediction result, accordingly, it is possible to occur total
Prediction result accuracy rate it is normal, but the relatively low situation of predictablity rate of some of loan product.In order to ensure to every
The predictablity rate of kind loan product, meets accuracy rate need of the technical staff to user's overdue prediction in different loan products
It asks, the loan product that can voluntarily select required analysis predictablity rate in the embodiment of the present invention by technical staff carries out targetedly
Analysis, and overdue prediction model is updated when accuracy rate is relatively low.
Corresponding to the method for foregoing embodiments, Fig. 5 shows that overdue prediction model provided in an embodiment of the present invention generates dress
The structure diagram set illustrates only and the relevant part of the embodiment of the present invention for convenience of description.The exemplary overdue predictions of Fig. 5
Model generating means can be the executive agent for the overdue prediction model generation method that previous embodiment one provides.
With reference to Fig. 5, which includes：
Data acquisition module 51, the L kinds user attribute data for obtaining M user and overdue class label, obtain
Dimension is the input matrix of M × (L+1), wherein M and L is positive integer.
Attribute selection module 52, for the input matrix to be input in preset learning model, filter out with it is described
The overdue maximum N kinds user attribute data of the class label degree of association, the learning model are to carry out user property and overdue classification
The learning model of label correlation analysis, wherein N is the positive integer less than or equal to L.
User grouping module 53 obtains number of users P for being grouped at random to the M userhH random use
Family group, and it is P to generate the corresponding dimension of the H user grouphThe H training sample matrix of × (N+1), wherein PhFor just
Integer, H are the positive integer more than 1, h ∈ [1, H].
Submodel builds module 54, for the H training sample matrix to be inputted preset H neural network mould respectively
It is trained in type up to meeting the preset condition of convergence, determines the corresponding model ginseng of the H neural network model
Number obtains H sub- prediction models.
Model construction module 55, the number of users P for including according to the H user grouphCalculate the H son prediction
The corresponding H ballot coefficient of model, and based on the H sub- prediction models and corresponding H ballot system
Number, structure obtain overdue prediction model：
Presult=Vcoent1Vresult1+Vcoent2Vresult2+…
+VcoenthVresulth+…+VcoentHVresultH
Wherein, Presult is the output of the overdue prediction model as a result, VresulthFor h-th of sub- prediction mould
The output of type is as a result, VcoenthFor the corresponding ballot coefficient of h-th of sub- prediction model.
Further, attribute selection module 52, including：
Comentropy computing module, for calculating the corresponding comentropy of the M user based on the overdue class label.
Information computational module is divided, for carrying out two dimensionization processing to the L kinds user attribute data, and calculates the L
The corresponding division information entropy of kind user attribute data and division information.
Information gain computing module, for calculating the M user based on described information entropy and the division information entropy
Corresponding information gain.
Information gain-ratio computing module is used for calculating the L kinds based on the division information and described information gain
The corresponding information gain-ratio of family attribute data.
Module is computed repeatedly, for randomly selecting out I to L-1 kind user attribute datas from the L kinds user attribute data,
And it is randomly selected based on describedGroup user attribute data, returns and is based on described in executing
The overdue class label calculates the operation of the corresponding comentropy of the M user, corresponding total to obtainA information gain-ratio, wherein I are the positive integer less than or equal to L-1.
User property determining module, it is described for finding outA information gain
The corresponding user attribute data of maximum described information ratio of profit increase in rate, to determine the N kinds user attribute data.
Further, user grouping module 53, including：
First user grouping module, if being greater than or equal to number of users threshold value for M, use can put back to formula or can not put back to
The methods of sampling of formula to carry out random sampling to the M user, to realize the random grouping to the M user, is wrapped
Containing the H user group that number of users is random.
Second user grouping module, if for M be less than the number of users threshold value, using the methods of sampling that can put back to formula come pair
The M user carries out random sampling, to realize the random grouping to the M user, obtains comprising the random institute of number of users
State H user group.
Further, the overdue prediction model generating means further include：
Accuracy rate detection module, the predictablity rate for obtaining the overdue prediction model of output in preset time period,
And judge whether the predictablity rate is less than accuracy rate threshold value.
Data update module obtains T if being that the predictablity rate is less than the accuracy rate threshold value for judging result
The N kinds user attribute data of user and overdue class label obtain the sample matrix that dimension is T × (N+1), wherein T
For positive integer.
User grouping module obtains number of users P for being grouped at random to the T userhH random user
Group, and it is P to generate the corresponding dimension of the H user grouphThe H training sample matrix of × (N+1), wherein PhIt is just whole
Number, H are the positive integer more than 1, h ∈ [1, H].
Model modification module, for returning, execution is described to input preset H god respectively by the H training sample matrix
Through being trained in network model up to meeting the preset condition of convergence, determine that the H neural network model corresponds to respectively
Model parameter, the operation of H sub- prediction models is obtained, to carry out model modification to the overdue prediction model.
Further, accuracy rate detection module, including：
Product type parsing module chooses the pointed loan product class of instruction for parsing loan product input by user
Type.
Product accuracy rate parsing module, for the overdue prediction model of output out of the preset time period that get
Predictablity rate in, parse the corresponding predictablity rate of the loan product type.
Accuracy rate contrast module, for judging whether the corresponding predictablity rate of the loan product type is less than the standard
True rate threshold value.
Each module realizes the process of respective function in overdue prediction model generating means provided in an embodiment of the present invention, specifically
The description of aforementioned embodiment illustrated in fig. 1 one is can refer to, details are not described herein again.
It should be understood that the size of the serial number of each step is not meant that the order of the execution order in above-described embodiment, each process
Execution sequence should be determined by its function and internal logic, the implementation process without coping with the embodiment of the present invention constitutes any limit
It is fixed.
Although will also be appreciated that term " first ", " second " etc. are used in some embodiment of the present invention in the text
Various elements are described, but these elements should not be limited by these terms.These terms are used only to an element
It is distinguished with another element.For example, the first contact can be named as the second contact, and similarly, the second contact can be by
It is named as the first contact, without departing from the range of various described embodiments.First contact and the second contact are all contacts, but
Be them it is not same contact.
Fig. 6 is the schematic diagram that the overdue prediction model that one embodiment of the invention provides generates terminal device.As shown in fig. 6,
The overdue prediction model of the embodiment generates terminal device 6：Processor 60, memory 61 store in the memory 61
There is the computer program 62 that can be run on the processor 60.The processor 60 is realized when executing the computer program 62
Step in above-mentioned each overdue prediction model generation method embodiment, such as step 101 shown in FIG. 1 is to 105.Alternatively, institute
The function that each module/unit in above-mentioned each device embodiment is realized when processor 60 executes the computer program 62 is stated, such as
The function of module 51 to 55 shown in Fig. 5.
It can be desktop PC, notebook, palm PC and high in the clouds that the overdue prediction model, which generates terminal device 6,
The computing devices such as server.The overdue prediction model generates terminal device and may include, but is not limited only to, processor 60, storage
Device 61.It will be understood by those skilled in the art that Fig. 6 is only the example that overdue prediction model generates terminal device 6, do not constitute
The restriction that terminal device 6 is generated to overdue prediction model may include components more more or fewer than diagram, or combine certain
Component or different components, such as it can also include input sending device, net that the overdue prediction model, which generates terminal device,
Network access device, bus etc..
Alleged processor 60 can be central processing unit (Central Processing Unit, CPU), can also be
Other general processors, digital signal processor (Digital Signal Processor, DSP), application-specific integrated circuit
(Application Specific Integrated Circuit, ASIC), ready-made programmable gate array (Field-
Programmable Gate Array, FPGA) either other programmable logic device, discrete gate or transistor logic,
Discrete hardware components etc..General processor can be microprocessor or the processor can also be any conventional processor
Deng.
The memory 61 can be that the overdue prediction model generates the internal storage unit of terminal device 6, such as exceedes
Phase prediction model generates the hard disk or memory of terminal device 6.The memory 61 can also be that the overdue prediction model generates
The External memory equipment of terminal device 6, such as the overdue prediction model generate the plug-in type hard disk being equipped on terminal device 6,
Intelligent memory card (Smart Media Card, SMC), secure digital (Secure Digital, SD) card, flash card (Flash
Card) etc..Further, the memory 61 can also both include the inside that the overdue prediction model generates terminal device 6
Storage unit also includes External memory equipment.The memory 61 is for storing the computer program and the overdue prediction
Model generates other programs and data needed for terminal device.The memory 61 can be also used for temporarily storing and send
Or the data that will be sent.
In addition, each functional unit in each embodiment of the present invention can be integrated in a processing unit, it can also
It is that each unit physically exists alone, it can also be during two or more units be integrated in one unit.Above-mentioned integrated list
The form that hardware had both may be used in member is realized, can also be realized in the form of SFU software functional unit.
If the integrated module/unit be realized in the form of SFU software functional unit and as independent product sale or
In use, can be stored in a computer read/write memory medium.Based on this understanding, the present invention realizes above-mentioned implementation
All or part of flow in example method, can also instruct relevant hardware to complete, the meter by computer program
Calculation machine program can be stored in a computer readable storage medium, the computer program when being executed by processor, it can be achieved that on
The step of stating each embodiment of the method.Wherein, the computer program includes computer program code, the computer program generation
Code can be source code form, object identification code form, executable file or certain intermediate forms etc..The computer-readable medium
May include：Any entity or device, recording medium, USB flash disk, mobile hard disk, magnetic of the computer program code can be carried
Dish, CD, computer storage, read-only memory (Read-Only Memory, ROM), random access memory (Random
Access Memory, RAM), electric carrier signal, telecommunication signal and software distribution medium etc..It should be noted that the meter
The content that calculation machine readable medium includes can carry out increase and decrease appropriate according to legislation in jurisdiction and the requirement of patent practice,
Such as in certain jurisdictions, according to legislation and patent practice, computer-readable medium does not include electric carrier signal and telecommunications
Signal.
Embodiment described above is merely illustrative of the technical solution of the present invention, rather than its limitations；Although with reference to aforementioned reality
Applying example, invention is explained in detail, it will be understood by those of ordinary skill in the art that：It still can be to aforementioned each
Technical solution recorded in embodiment is modified or equivalent replacement of some of the technical features；And these are changed
Or replace, so that the essence of corresponding technical solution is detached from the spirit and scope of various embodiments of the present invention technical solution, it should all
It is included within protection scope of the present invention.
Claims (10)
1. a kind of overdue prediction model generation method, which is characterized in that including：
The L kinds user attribute data of M user of acquisition and overdue class label obtain the input square that dimension is M × (L+1)
Battle array, wherein M and L is positive integer；
The input matrix is input in preset learning model, is filtered out maximum with the overdue class label degree of association
N kind user attribute datas, the learning model are the learning model for carrying out user property and overdue class label correlation analysis,
Wherein, N is the positive integer less than or equal to L；
The M user is grouped at random, obtains number of users PhH random user group, and generate the H user group
Corresponding dimension is PhThe H training sample matrix of × (N+1), wherein PhFor positive integer, H is the positive integer more than 1, h
∈ [1, H]；
The H training sample matrix is inputted respectively in preset H neural network model and is trained until meeting default
The condition of convergence, determine the corresponding model parameter of the H neural network model, obtain H sub- prediction models；
The number of users P for including according to the H user grouphCalculate the H sub- prediction model corresponding H ballot systems
Number, and obtain overdue prediction model based on the H sub- prediction models and the corresponding H ballot coefficient, structure：
Presult=Vcoent1Vresult1+Vcoent2Vresult2+…
+VcoenthVresulth+…+VcoentHVresultH
Wherein, Presult is the output of the overdue prediction model as a result, VresulthFor the defeated of h-th sub- prediction model
Go out as a result, VcoenthFor the corresponding ballot coefficient of h-th of sub- prediction model.
2. overdue prediction model generation method as described in claim 1, which is characterized in that the processing procedure of the learning model
Including：
The corresponding comentropy of the M user is calculated based on the overdue class label；
Two dimensionization processing is carried out to the L kinds user attribute data, and calculates the corresponding division letter of the L kinds user attribute data
Cease entropy and division information；
The corresponding information gain of the M user is calculated based on described information entropy and the division information entropy；
The corresponding information gain-ratio of the L kinds user attribute data is calculated based on the division information and described information gain；
I is randomly selected out to L-1 kind user attribute datas from the L kinds user attribute data, and is randomly selected based on described
It arrivesGroup user attribute data is returned and is calculated based on the overdue class label described in executing
The operation of the corresponding comentropy of the M user, it is corresponding total to obtainA information
Ratio of profit increase, wherein I are the positive integer less than or equal to L-1；
It finds out describedMaximum described information ratio of profit increase pair in a information gain-ratio
The user attribute data answered, to determine the N kinds user attribute data.
3. overdue prediction model generation method as described in claim 1, which is characterized in that described to be carried out to the M user
Random grouping, obtains number of users PhH random user group, including：
If M is greater than or equal to number of users threshold value, described M is used using that can put back to formula or the methods of sampling of formula can not be put back to
Family carries out random sampling, to realize the random grouping to the M user, obtains comprising the random H user of number of users
Group；
If M is less than the number of users threshold value, random sampling is carried out to the M user using the methods of sampling that can put back to formula,
To realize the random grouping to the M user, obtain comprising the random H user group of number of users.
4. overdue prediction model generation method as described in claim 1, which is characterized in that obtain overdue prediction model described
Later, further include：
The predictablity rate of the overdue prediction model of output in preset time period is obtained, and whether judges the predictablity rate
Less than accuracy rate threshold value；
If judging result, which is the predictablity rate, is less than the accuracy rate threshold value, the N kinds user property of T user is obtained
Data and overdue class label obtain the sample matrix that dimension is T × (N+1), wherein T is positive integer；
The T user is grouped at random, obtains number of users PhH random user group, and generate the H user group
Corresponding dimension is PhThe H training sample matrix of × (N+1), wherein PhFor positive integer, H is the positive integer more than 1, h
∈ [1, H]；
Described input the H training sample matrix in preset H neural network model respectively of execution is returned to be trained
Until meeting the preset condition of convergence, the corresponding model parameter of the H neural network model is determined, obtain H son
The operation of prediction model, to carry out model modification to the overdue prediction model.
5. overdue prediction model generation method as claimed in claim 4, which is characterized in that institute in the acquisition preset time period
The predictablity rate for exporting overdue prediction model is stated, and judges whether the predictablity rate is less than accuracy rate threshold value, including：
It parses loan product input by user and chooses the pointed loan product type of instruction；
From the predictablity rate of the overdue prediction model of output in the preset time period got, described borrow is parsed
Borrow the corresponding predictablity rate of product type；
Judge whether the corresponding predictablity rate of the loan product type is less than the accuracy rate threshold value.
6. a kind of overdue prediction model generates terminal device, which is characterized in that the overdue prediction model generates processing terminal and sets
Standby includes memory, processor, and the computer program that can be run on the processor, the place are stored on the memory
Reason device realizes following steps when executing the computer program：
The L kinds user attribute data of M user of acquisition and overdue class label obtain the input square that dimension is M × (L+1)
Battle array, wherein M and L is positive integer；
The input matrix is input in preset learning model, is filtered out maximum with the overdue class label degree of association
N kind user attribute datas, the learning model are the learning model for carrying out user property and overdue class label correlation analysis,
Wherein, N is the positive integer less than or equal to L；
The M user is grouped at random, obtains number of users PhH random user group, and generate the H user group
Corresponding dimension is PhThe H training sample matrix of × (N+1), wherein PhFor positive integer, H is the positive integer more than 1, h
∈ [1, H]；
The H training sample matrix is inputted respectively in preset H neural network model and is trained until meeting default
The condition of convergence, determine the corresponding model parameter of the H neural network model, obtain H sub- prediction models；
The number of users P for including according to the H user grouphCalculate the H sub- prediction model corresponding H ballot systems
Number, and obtain overdue prediction model based on the H sub- prediction models and the corresponding H ballot coefficient, structure：
Presult=Vcoent1Vresult1+Vcoent2Vresult2+…
+VcoenthVresulth+…+VcoentHVresultH
Wherein, Presult is the output of the overdue prediction model as a result, VresulthFor the defeated of h-th sub- prediction model
Go out as a result, VcoenthFor the corresponding ballot coefficient of h-th of sub- prediction model.
7. overdue prediction model generates terminal device as claimed in claim 6, which is characterized in that the learning model processes
Journey specifically includes：
The corresponding comentropy of the M user is calculated based on the overdue class label；
Two dimensionization processing is carried out to the L kinds user attribute data, and calculates the corresponding division letter of the L kinds user attribute data
Cease entropy and division information；
The corresponding information gain of the M user is calculated based on described information entropy and the division information entropy；
The corresponding information gain-ratio of the L kinds user attribute data is calculated based on the division information and described information gain；
I is randomly selected out to L-1 kind user attribute datas from the L kinds user attribute data, and is randomly selected based on described
It arrivesGroup user attribute data is returned and is calculated based on the overdue class label described in executing
The operation of the corresponding comentropy of the M user, it is corresponding total to obtainA information
Ratio of profit increase, wherein I are the positive integer less than or equal to L-1；
It finds out describedMaximum described information ratio of profit increase pair in a information gain-ratio
The user attribute data answered, to determine the N kinds user attribute data.
8. overdue prediction model generates terminal device as claimed in claim 6, which is characterized in that it is described to the M user into
The random grouping of row, obtains number of users PhH random user group, specifically includes：
If M is greater than or equal to number of users threshold value, described M is used using that can put back to formula or the methods of sampling of formula can not be put back to
Family carries out random sampling, to realize the random grouping to the M user, obtains comprising the random H user of number of users
Group；
If M is less than the number of users threshold value, random sampling is carried out to the M user using the methods of sampling that can put back to formula,
To realize the random grouping to the M user, obtain comprising the random H user group of number of users.
9. overdue prediction model generates terminal device as claimed in claim 6, which is characterized in that the processor executes the meter
Following steps are also realized when calculation machine program：
The predictablity rate of the overdue prediction model of output in preset time period is obtained, and whether judges the predictablity rate
Less than accuracy rate threshold value；
If judging result, which is the predictablity rate, is less than the accuracy rate threshold value, the N kinds user property of T user is obtained
Data and overdue class label obtain the sample matrix that dimension is T × (N+1), wherein T is positive integer；
The T user is grouped at random, obtains number of users PhH random user group, and generate the H user group
Corresponding dimension is PhThe H training sample matrix of × (N+1), wherein PhFor positive integer, H is the positive integer more than 1, h
∈ [1, H]；
Described input the H training sample matrix in preset H neural network model respectively of execution is returned to be trained
Until meeting the preset condition of convergence, the corresponding model parameter of the H neural network model is determined, obtain H son
The operation of prediction model, to carry out model modification to the overdue prediction model.
10. a kind of computer readable storage medium, the computer-readable recording medium storage has computer program, feature to exist
In when the computer program is executed by processor the step of any one of such as claim 1 to 5 of realization the method.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810436131.2A CN108573358A (en) | 2018-05-09 | 2018-05-09 | A kind of overdue prediction model generation method and terminal device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810436131.2A CN108573358A (en) | 2018-05-09 | 2018-05-09 | A kind of overdue prediction model generation method and terminal device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108573358A true CN108573358A (en) | 2018-09-25 |
Family
ID=63572027
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810436131.2A Pending CN108573358A (en) | 2018-05-09 | 2018-05-09 | A kind of overdue prediction model generation method and terminal device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108573358A (en) |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109558674A (en) * | 2018-11-28 | 2019-04-02 | 北京超萌国际文化有限公司 | Method for Sales Forecast and its model training method, device |
| CN110246026A (en) * | 2019-05-21 | 2019-09-17 | 平安银行股份有限公司 | A kind of output combination setting method, device and the terminal device of data transfer |
| CN110349038A (en) * | 2019-06-13 | 2019-10-18 | 中国平安人寿保险股份有限公司 | Risk evaluation model training method and methods of risk assessment |
| WO2020119096A1 (en) * | 2018-12-13 | 2020-06-18 | 平安医疗健康管理股份有限公司 | Data analysis-based hospital evaluation method and related product |
| CN111311397A (en) * | 2020-02-13 | 2020-06-19 | 上海凯岸信息科技有限公司 | Scheme for improving voice call-out robot collection fee collection rate by combining scoring card model and ABtest |
| CN111767982A (en) * | 2020-05-20 | 2020-10-13 | 北京大米科技有限公司 | Training method and device for user conversion prediction model, storage medium and electronic equipment |
| WO2020253038A1 (en) * | 2019-06-18 | 2020-12-24 | 平安普惠企业管理有限公司 | Model construction method and apparatus |
| CN112348583A (en) * | 2020-11-04 | 2021-02-09 | 贝壳技术有限公司 | User preference generation method and generation system |
| CN112656401A (en) * | 2019-10-15 | 2021-04-16 | 梅州市青塘实业有限公司 | Intelligent monitoring method, device and equipment |
| CN112789636A (en) * | 2018-10-26 | 2021-05-11 | 索尼公司 | Information processing apparatus, information processing method, and program |
-
2018
- 2018-05-09 CN CN201810436131.2A patent/CN108573358A/en active Pending
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112789636A (en) * | 2018-10-26 | 2021-05-11 | 索尼公司 | Information processing apparatus, information processing method, and program |
| CN109558674A (en) * | 2018-11-28 | 2019-04-02 | 北京超萌国际文化有限公司 | Method for Sales Forecast and its model training method, device |
| WO2020119096A1 (en) * | 2018-12-13 | 2020-06-18 | 平安医疗健康管理股份有限公司 | Data analysis-based hospital evaluation method and related product |
| CN110246026A (en) * | 2019-05-21 | 2019-09-17 | 平安银行股份有限公司 | A kind of output combination setting method, device and the terminal device of data transfer |
| CN110246026B (en) * | 2019-05-21 | 2023-06-27 | 平安银行股份有限公司 | Data transfer output combination setting method and device and terminal equipment |
| CN110349038A (en) * | 2019-06-13 | 2019-10-18 | 中国平安人寿保险股份有限公司 | Risk evaluation model training method and methods of risk assessment |
| WO2020253038A1 (en) * | 2019-06-18 | 2020-12-24 | 平安普惠企业管理有限公司 | Model construction method and apparatus |
| CN112656401A (en) * | 2019-10-15 | 2021-04-16 | 梅州市青塘实业有限公司 | Intelligent monitoring method, device and equipment |
| CN112656401B (en) * | 2019-10-15 | 2023-08-22 | 梅州市青塘实业有限公司 | Intelligent monitoring method, device and equipment |
| CN111311397A (en) * | 2020-02-13 | 2020-06-19 | 上海凯岸信息科技有限公司 | Scheme for improving voice call-out robot collection fee collection rate by combining scoring card model and ABtest |
| CN111767982A (en) * | 2020-05-20 | 2020-10-13 | 北京大米科技有限公司 | Training method and device for user conversion prediction model, storage medium and electronic equipment |
| CN112348583A (en) * | 2020-11-04 | 2021-02-09 | 贝壳技术有限公司 | User preference generation method and generation system |
| CN112348583B (en) * | 2020-11-04 | 2022-12-06 | 贝壳技术有限公司 | User preference generation method and generation system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108573358A (en) | A kind of overdue prediction model generation method and terminal device | |
| TWI788529B (en) | Credit risk prediction method and device based on LSTM model | |
| Whittle | Probability via expectation | |
| Wang et al. | An improved boosting based on feature selection for corporate bankruptcy prediction | |
| US20110153536A1 (en) | Computer-Implemented Systems And Methods For Dynamic Model Switching Simulation Of Risk Factors | |
| CN110264270B (en) | Behavior prediction method, behavior prediction device, behavior prediction equipment and storage medium | |
| DE112020002684T5 (en) | A multi-process system for optimal predictive model selection | |
| CN110135701A (en) | Control automatic generation method, device, electronic equipment and the readable medium of rule | |
| CN113706291A (en) | Fraud risk prediction method, device, equipment and storage medium | |
| CN111507461B (en) | Interpretability information determining method and device | |
| CN111325572B (en) | Data processing method and device | |
| CN107844874A (en) | Enterprise operation problem analysis system and its method | |
| CN114298825A (en) | Method and device for extremely evaluating repayment volume | |
| CN113420789A (en) | Method, device, storage medium and computer equipment for predicting risk account | |
| Goodman | Uncertainty, risk, and decision: the PVA example | |
| CN113435748A (en) | Dot state determination method and apparatus, electronic device and storage medium | |
| CN112308294A (en) | Default probability prediction method and device | |
| CN111161042A (en) | Personal risk assessment method, device, terminal and storage medium | |
| CN110348190A (en) | User equipment ownership judgment method and device based on user's operation behavior | |
| CN110362981A (en) | The method and system of abnormal behaviour are judged based on credible equipment fingerprint | |
| CN116484230B (en) | Method for identifying abnormal business data and training method of AI digital person | |
| CN117815674B (en) | Game information recommendation method and device, computer readable medium and electronic equipment | |
| CN116340765B (en) | Electricity larceny user prediction method and device, storage medium and electronic equipment | |
| Zaychenko et al. | Banks Financial State Analysis and Bankruptcy Risk Forecasting with Application of Fuzzy Neural Networks | |
| Aslan | Assessment of Post-Crisis Financial Performance and Actions in Italian Companies |
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 |