CN113240512A - Method, device, readable medium and equipment for constructing risk prediction model - Google Patents
Method, device, readable medium and equipment for constructing risk prediction model Download PDFInfo
- Publication number
- CN113240512A CN113240512A CN202110661850.6A CN202110661850A CN113240512A CN 113240512 A CN113240512 A CN 113240512A CN 202110661850 A CN202110661850 A CN 202110661850A CN 113240512 A CN113240512 A CN 113240512A
- Authority
- CN
- China
- Prior art keywords
- model
- machine learning
- client
- participant node
- learning 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
- 238000013058 risk prediction model Methods 0.000 title claims abstract description 139
- 238000000034 method Methods 0.000 title claims abstract description 60
- 238000010801 machine learning Methods 0.000 claims abstract description 210
- 238000012549 training Methods 0.000 claims abstract description 167
- 230000008520 organization Effects 0.000 claims description 27
- 238000010276 construction Methods 0.000 claims description 9
- 230000006399 behavior Effects 0.000 claims description 7
- 238000004590 computer program Methods 0.000 claims description 3
- 230000008569 process Effects 0.000 description 11
- 230000009471 action Effects 0.000 description 3
- 238000012502 risk assessment Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 2
- 238000011835 investigation Methods 0.000 description 2
- 238000012546 transfer Methods 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000012417 linear regression Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000007637 random forest analysis Methods 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
Images
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
- G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
- G06Q40/03—Credit; Loans; Processing thereof
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
- G06Q10/10—Office automation; Time management
- G06Q10/103—Workflow collaboration or project management
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- Strategic Management (AREA)
- Human Resources & Organizations (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Economics (AREA)
- Entrepreneurship & Innovation (AREA)
- General Business, Economics & Management (AREA)
- Marketing (AREA)
- Development Economics (AREA)
- Software Systems (AREA)
- Accounting & Taxation (AREA)
- Finance (AREA)
- Evolutionary Computation (AREA)
- Tourism & Hospitality (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Quality & Reliability (AREA)
- Operations Research (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Medical Informatics (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Technology Law (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Game Theory and Decision Science (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Financial Or Insurance-Related Operations Such As Payment And Settlement (AREA)
Abstract
The application discloses a method, a device, a readable medium and equipment for constructing a risk prediction model, wherein the method comprises the steps of acquiring credit characteristic data and actual overdue condition data of each local client; training a local initial machine learning model to obtain a trained machine learning model; and if the trained machine learning model does not meet the training cutoff condition, sending the model parameters of the trained machine learning model to the next participant node, and determining the model meeting the training cutoff condition as the loan risk prediction model until the participant node in the alliance trains to obtain the machine learning model meeting the training cutoff condition. The loan risk prediction model is obtained through local client sample data training of a plurality of participant nodes, and only model parameters are transmitted among the participant nodes without transmitting local client data, so that a high-quality model is constructed while the safety and privacy of data are guaranteed.
Description
Technical Field
The present application relates to the field of computer technologies, and in particular, to a method, an apparatus, a readable medium, and a device for constructing a risk prediction model.
Background
In the prior art, when a client of a financial institution applies for a loan from the financial institution, the financial institution usually performs a risk assessment according to the credit status of the user and then decides whether to loan the client according to the risk assessment result. Specifically, in the existing risk assessment method, a client wind control model is mainly constructed by using the historical consumption behaviors and personal credit investigation of a client and based on statistical methods such as linear regression and the like, then the constructed client wind control model is used for predicting the loan overdue condition of the client, and finally whether the client is loaned or not is determined according to the predicted loan overdue condition.
However, each financial institution needs to ensure the security of client data of the financial institution and the privacy of the client, so that the financial institution can only select the client data of the financial institution to perform model training in the process of constructing the client wind control model, which results in that the client sample data used by the client wind control model in the training process is relatively single, the client wind control model finally trained by the financial institution is not good enough, and the loan overdue condition of the client cannot be accurately predicted.
Disclosure of Invention
Based on the defects of the prior art, the application provides a method, a device, a readable medium and equipment for constructing a risk prediction model, so that a high-quality risk prediction model is trained by using abundant client sample data under the condition of ensuring data security and client privacy.
The first aspect of the application discloses a method for constructing a risk prediction model, which is applied to target participant nodes, wherein the target participant nodes refer to each participant node in a federal organization; the federal agency is composed of each of the participant nodes connected in series in a specific order; the construction method of the risk prediction model comprises the following steps:
acquiring credit characteristic data of each local client and actual overdue condition data of each client; the actual overdue condition data of the client is used for explaining whether the client actually has overdue repayment behavior at a historical time point;
training a local initial machine learning model by using credit characteristic data of each client and actual overdue condition data of each client to obtain a machine learning model after the target participant node is trained; if the target participant node is the first participant node to be trained in the federal agency, the model parameters of the local initial machine model are initial model parameters; if the target participant node is not the participant node of the first training model in the federal agency, the model parameters of the local initial machine model are the model parameters of the machine learning model trained by the previous participant node;
if the machine learning model trained by the target participant node meets a training cutoff condition, determining the machine learning model trained by the target participant node as a loan risk prediction model;
if the machine learning model trained by the target participant node does not meet the training cutoff condition, sending model parameters of the machine learning model trained by the target participant node to a next participant node until the participant node in the alliance trains to obtain the machine learning model meeting the training cutoff condition, and determining the machine learning model meeting the training cutoff condition as the loan risk prediction model; the loan risk prediction model is a model used by a participant node in the federal agency to predict the overdue condition of the pending clients of the participant node.
Optionally, in the method for constructing a risk prediction model, the training cutoff condition is: the model of the trained machine learning model is converged in the federal agency, the training times of the trained machine learning model in the federal agency are larger than or equal to a threshold value of the training times, or the training time of the trained machine learning model in the federal agency is larger than or equal to a time threshold value.
Optionally, in the method for constructing a risk prediction model, the training a local initial machine learning model by using the credit feature data of each client and the actual overdue condition data of each client to obtain the machine learning model trained by the target participant node includes:
inputting the credit characteristic data of each customer into a local initial machine learning model, and outputting the predicted overdue condition data of each customer by the local initial machine learning model; the predicted overdue condition data of the client is used for explaining whether the client is overdue and repayment predicted by the local initial machine learning model;
and continuously adjusting model parameters in the local initial machine learning model by using the error between the predicted overdue condition data of each client and the actual overdue condition data of the client until the error between the predicted overdue condition data of each client and the actual overdue condition data of the client output by the adjusted local initial machine learning model meets a preset convergence condition, and determining the adjusted local initial machine learning model as the machine learning model trained by the target participant node.
Optionally, in the method for constructing a risk prediction model, after determining the machine learning model trained by the target participant node as a loan risk prediction model, the method further includes:
acquiring credit characteristic data of a client to be examined and approved;
inputting the credit characteristic data of the client to be approved into the loan risk prediction model, and outputting the predicted overdue condition data of the client to be approved by the loan risk prediction model; the predicted overdue condition data of the client to be approved is used for explaining whether the loan risk prediction model predicts the overdue repayment of the client to be approved;
if the predicted overdue condition data of the client to be checked shows that the loan risk prediction model predicts the overdue repayment of the client to be checked, rejecting the loan application of the client to be checked;
and if the predicted overdue condition data of the client to be checked shows that the loan risk prediction model predicts that the client to be checked will not be overdue for repayment, accepting the loan application of the client to be checked.
Optionally, in the above method for constructing a risk prediction model, the data type of the credit feature data acquired by the target participant node is consistent with the data type of the credit feature data acquired by each of the participant nodes in the federal organization.
Optionally, in the method for constructing a risk prediction model, after determining that the machine learning model trained by the target participant node is the loan risk prediction model if the machine learning model trained by the target participant node satisfies a training cutoff condition, the method further includes:
and sending the model parameters of the loan risk prediction model to each participant node in the federal organization, and respectively constructing the loan risk prediction model by each participant node by using the model parameters of the loan risk prediction model.
Optionally, in the above method for constructing a risk prediction model, the credit feature data of the client includes: at least one of transaction data for the customer, asset data for the customer, and blacklist record data for the customer.
The second aspect of the application discloses a risk prediction model construction device, which is applied to target participant nodes, wherein the target participant nodes refer to each participant node in a federal organization; the federal agency is composed of each of the participant nodes connected in series in a specific order; the risk prediction model construction device comprises:
the system comprises a first acquisition unit, a second acquisition unit and a third acquisition unit, wherein the first acquisition unit is used for acquiring credit characteristic data of each local client and actual overdue condition data of each client; the actual overdue condition data of the client is used for explaining whether the client actually has overdue repayment behavior at a historical time point;
the training unit is used for training a local initial machine learning model by using the credit characteristic data of each client and the actual overdue condition data of each client to obtain the machine learning model after the node of the target participant is trained; if the target participant node is the first participant node to be trained in the federal agency, the model parameters of the local initial machine model are initial model parameters; if the target participant node is not the participant node of the first training model in the federal agency, the model parameters of the local initial machine model are the model parameters of the machine learning model trained by the previous participant node;
the first determining unit is used for determining the machine learning model trained by the target participant node as a loan risk prediction model if the machine learning model trained by the target participant node meets a training cutoff condition;
a first sending unit, configured to send, if the machine learning model trained by the target participant node does not meet the training cutoff condition, a model parameter of the machine learning model trained by the target participant node to a next participant node, and determine, until there is a machine learning model meeting the training cutoff condition in the coalition enterprises, the machine learning model meeting the training cutoff condition as the loan risk prediction model when the participant node is trained to obtain the machine learning model meeting the training cutoff condition; the loan risk prediction model is a model used by a participant node in the federal agency to predict the overdue condition of the pending clients of the participant node.
Optionally, in the above apparatus for constructing a risk prediction model, the training cutoff condition is: the model of the trained machine learning model is converged in the federal agency, the training times of the trained machine learning model in the federal agency are larger than or equal to a threshold value of the training times, or the training time of the trained machine learning model in the federal agency is larger than or equal to a time threshold value.
Optionally, in the above apparatus for constructing a risk prediction model, the training unit includes:
the first input subunit is used for inputting the credit characteristic data of each client into a local initial machine learning model, and outputting the predicted overdue condition data of each client by the local initial machine learning model; the predicted overdue condition data of the client is used for explaining whether the client is overdue and repayment predicted by the local initial machine learning model;
and the adjusting subunit is configured to continuously adjust the model parameters in the local initial machine learning model by using an error between the predicted overdue condition data of each client and the actual overdue condition data of the client until an error between the predicted overdue condition data of each client output by the adjusted local initial machine learning model and the actual overdue condition data of the client meets a preset convergence condition, and determine the adjusted local initial machine learning model as the machine learning model trained by the target participant node.
Optionally, in the above apparatus for constructing a risk prediction model, the method further includes:
the second acquisition unit is used for acquiring credit characteristic data of the client to be approved;
the second input subunit is used for inputting the credit characteristic data of the client to be approved to the loan risk prediction model and outputting the predicted overdue condition data of the client to be approved by the loan risk prediction model; the predicted overdue condition data of the client to be approved is used for explaining whether the loan risk prediction model predicts the overdue repayment of the client to be approved;
a refund subunit, configured to refund the loan application of the pending client if the predicted overdue condition data of the pending client indicates that the loan risk prediction model predicts the overdue repayment of the pending client;
and the acceptance subunit is used for accepting the loan application of the client to be checked if the predicted overdue condition data of the client to be checked shows that the loan risk prediction model predicts that the client to be checked will not pay overdue.
Optionally, in the risk prediction model building apparatus, a data type of the credit feature data acquired by the target participant node coincides with a data type of the credit feature data acquired by each of the participant nodes in the federal organization.
Optionally, in the above apparatus for constructing a risk prediction model, the method further includes:
and the second sending unit is used for sending the model parameters of the loan risk prediction model to each participant node in the federal organization, and each participant node constructs the loan risk prediction model by using the model parameters of the loan risk prediction model.
Optionally, in the above apparatus for constructing a risk prediction model, the credit feature data of the client includes: at least one of transaction data for the customer, asset data for the customer, and blacklist record data for the customer.
A third aspect of the application discloses a computer readable medium having a computer program stored thereon, wherein the program when executed by a processor implements the method as described in any of the first aspects above.
The fourth aspect of the present application discloses an apparatus comprising:
one or more processors;
a storage device having one or more programs stored thereon;
the one or more programs, when executed by the one or more processors, cause the one or more processors to implement the method as in any one of the first aspects above.
According to the technical scheme, the method for constructing the risk prediction model is applied to the target participant node, the target participant node refers to each participant node in a federal organization, the federal organization is composed of the participant nodes which are connected in series according to a specific sequence, the credit feature data of each local client and the actual overdue condition data of each client are obtained, then the credit feature data of each client and the actual overdue condition data of each client are utilized to train the local initial machine learning model, and the machine learning model after the target participant node is trained is obtained. If the target participant node is the first participant node to be trained in the federal agency, the model parameters of the local initial machine model are the initial model parameters, and if the target participant node is not the participant node of the first training model in the federal agency, the model parameters of the local initial machine model are the model parameters of the machine learning model trained by the previous participant node. And if the machine learning model after the target participant node training meets the training cutoff condition, determining the machine learning model after the target participant node training as a loan risk prediction model. And if the machine learning model after the target participant node training does not meet the training cutoff condition, sending the model parameters of the machine learning model after the target participant node training to the next participant node, and determining the machine learning model meeting the training cutoff condition as the loan risk prediction model until the participant node training in the alliance obtains the machine learning model meeting the training cutoff condition. The loan risk prediction model in the embodiment of the application is obtained by training the credit characteristic data of each client local to each participant node and the actual overdue condition data of each client, so that the used client sample data is richer than that in the prior art, and only model parameters of the machine learning model after the training of the participant nodes are transmitted among the participant nodes, but local client data is not transmitted, so that the security of local data and the privacy of the clients are guaranteed, and a high-quality model is constructed.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the provided drawings without creative efforts.
Fig. 1 is a schematic flowchart of a method for constructing a risk prediction model according to an embodiment of the present disclosure;
fig. 2 is a schematic structural diagram of a joint mechanism according to an embodiment of the present disclosure;
fig. 3 is a schematic flowchart of a training method of a machine learning model according to an embodiment of the present disclosure;
FIG. 4 is a flow chart illustrating a method for processing a loan application according to an embodiment of the invention;
fig. 5 is a schematic structural diagram of a risk prediction model construction apparatus according to an embodiment of the present application.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Referring to fig. 1, an embodiment of the present application provides a method for constructing a risk prediction model, where the method is applied to target participant nodes, where a target participant node refers to each participant node in a federal organization, and the federal organization is composed of each participant node connected in series according to a specific sequence, and the method for constructing a risk prediction model specifically includes the following steps:
s101, acquiring credit characteristic data of each local client and actual overdue condition data of each client.
The actual overdue condition data of the client is used for explaining whether the client actually has overdue payment behaviors at the historical time point. Specifically, a client database may be locally pre-constructed, and credit characteristic data of each client and actual overdue condition data of each client in the target participant node are stored in the client database, and the current credit characteristic data of each client and actual overdue condition data of each client may be collected and stored in the client database in real time, or the credit characteristic data of each client and actual overdue condition data of each client in each period may be collected according to a preset period. A wide table of attributes may be built in the customer database with customer representation as key values, customer credit profile data and actual overdue data.
When step S101 is executed, credit feature data of each client local to the target participant node and actual overdue condition data of each client may be obtained from a pre-constructed client database. Credit profile data is data used to characterize the credit status of a target user, such as assets, whether there is entry into a blacklist, transaction status, etc. The more assets, the higher the credit rating of the client, the lower the credit rating of the client entering the blacklist, and the transaction data can be consumption, repayment, credit investigation, credit card bill, transfer and the like, and can also reflect the credit condition of the user. The credit characteristic data of the client is history data, and may specifically be the credit characteristic data of the client at a history time point.
Step S101 may be executed when receiving the model parameters of the machine learning model trained by the previous participant node sent by the previous participant node, or when receiving an initial machine learning model for starting to train the local node, step S101 is executed. Specifically, referring to fig. 2, the federal agency is an agency that forms a closed loop ring link by connecting a plurality of participant nodes 201 in series in a specific order. Each participant node 201 in the federal agency executes the method of constructing the risk prediction model shown in fig. 1. The federal agency in the embodiment of the application is used for training a loan risk prediction model for predicting overdue conditions of pending clients of the participant nodes. The loan risk prediction model is applied to each participant node, and the participant nodes can predict the overdue condition of the client to be checked (namely whether the client is overdue) by using the loan risk prediction model, and further determine whether to accept the loan for the client according to the predicted result.
Specifically, for each participant node shown in fig. 2, after the participant node receives the model parameters of the machine learning model trained by the previous participant node (because a specific sequence is defined between the serially connected participant nodes, the participant node has a unique previous participant node), which are sent by the previous participant node, the machine learning model is constructed by using the received model parameters, and is used as a local initial machine learning model, and then step S101 is performed. If the participant node is the first node in the federal agency to start model training, that is, the model parameter sent by the previous participant node is not received, the initial model parameter is used to construct a local initial machine model, and then step S101 is started to be executed, where the initial model parameter may be a random value.
Optionally, in a specific embodiment of the present application, the data type of the credit feature data acquired by the target participant node is consistent with the data type of the credit feature data acquired by each participant node in the federal organization. In the embodiment of the application, a loan risk prediction model is trained by all the participant nodes in the federal institution together, so that the data types of the acquired credit feature data are consistent before each participant node performs training, and the training of a plurality of participant nodes is ensured to be the same model. For example, if the credit profile data obtained by the target participant node includes both asset data and transaction data, then the credit profile data obtained by all other participant nodes in the federal organization are also both asset data and transaction data.
Optionally, in an embodiment of the present application, the credit feature data of the client includes: at least one of transaction data for the customer, asset data for the customer, and blacklist record data for the customer.
S102, training the local initial machine learning model by using the credit characteristic data of each client and the actual overdue condition data of each client to obtain the machine learning model after the target participant node is trained.
If the target participant node is the first participant node to be trained in the federal agency, the model parameters of the local initial machine model are the initial model parameters, and if the target participant node is not the participant node of the first training model in the federal agency, the model parameters of the local initial machine model are the model parameters of the machine learning model trained by the previous participant node.
The local initial machine learning model is a supervised model, and specifically can be a model adopting algorithms such as random forest, extreme gradient lifting, gradient lifting tree, lightweight gradient lifting machine, neural network and the like. The actual overdue condition data of each client are input into the local initial machine learning model, and then model parameters of the local initial machine learning model are continuously adjusted, so that the predicted overdue condition data of each client and the actual overdue condition data of each client are continuously close to each other and are output by the local initial machine learning model, and the machine learning model after the node of the target participant is trained is obtained.
It should be noted that, when the target participant node is a first participant node in the federal agency for training, the model parameters of the local initial machine model in step S102 are initial model parameters, and when the target participant node is not a participant node of a first training model in the federal agency, the model parameters of the local initial machine model are model parameters of a machine learning model trained by a previous participant node, and the local initial machine model can be constructed by receiving the model parameters of the machine learning model trained by the previous participant node and then using the model parameters. The process of the last participant node obtaining the trained machine learning model is the same as step S102.
In this embodiment, when the model parameter of the local initial machine model used by the target participant node is the model parameter of the machine learning model trained by the previous participant node, the machine learning model trained by the target participant node obtained in step S102 is trained by using the credit feature data of the local client of the target participant node and the actual overdue condition data of each client, and the credit feature data of the local client of each participant node before the target participant node training model and the actual overdue condition data, that is, the machine learning model trained by the target participant node is not actually trained by only the local data.
Optionally, referring to fig. 3, in an embodiment of the present application, an implementation of step S102 is performed, including:
s301, inputting the credit characteristic data of each client into a local initial machine learning model, and outputting the predicted overdue condition data of each client by the local initial machine learning model.
The predicted overdue condition data of the client is used for explaining whether the client is overdue and paid or not predicted by the local initial machine learning model. The predicted overdue condition data of the client can be a probability value for predicting overdue payment of the client or a numerical value representing whether the client is overdue payment or not. For example, if the predicted overdue condition data of the client is 1, the client represents that the payment is overdue, and if the predicted overdue condition data of the client is 0, the client represents that the payment is not overdue.
In step S301, for each client, the credit feature data of the client is input into the local initial machine learning model, and the predicted overdue condition data of the client is output by the local initial machine learning model.
S302, continuously adjusting model parameters in the local initial machine learning model by using errors between the predicted overdue condition data of each client and the actual overdue condition data of the client until the errors between the predicted overdue condition data of each client and the actual overdue condition data of the client output by the adjusted local initial machine learning model meet preset convergence conditions, and determining the adjusted local initial machine learning model as the machine learning model after the target participant node is trained.
Specifically, for each customer, an error between the customer's predicted overdue condition data and the customer's actual overdue condition data is calculated. And continuously adjusting model parameters in the local initial machine learning model by using the error between the predicted overdue condition data of each client and the actual overdue condition data of the client, so that the predicted overdue condition data of the client and the actual overdue condition data of the client are continuously close to each other until the error between the predicted overdue condition data of each client and the actual overdue condition data of the client output by the adjusted local initial machine learning model meets a preset convergence condition, and determining the adjusted local initial machine learning model as the machine learning model trained by the target participant node.
S103, judging whether the machine learning model after the target participant node training meets a training cutoff condition.
If the machine learning model after the target participant node training does not meet the training cutoff condition, it indicates that the machine learning model after the target participant node training still needs to be continuously trained and optimized, so step S105 needs to be executed, and when the machine learning model after the target participant node training meets the training cutoff condition, it indicates that the machine learning model after the target participant node training does not need to be continuously trained and optimized in the federal agency.
Optionally, in a specific embodiment of the present application, the training cutoff condition is: the model of the trained machine learning model is converged in the federal agency, the training times of the trained machine learning model in the federal agency are larger than or equal to a threshold value of the training times, or the training time of the trained machine learning model in the federal agency is larger than or equal to a time threshold value.
Specifically, when the training cutoff condition is that the model of the trained machine learning model converges in the federal agency, the process of executing step S103 is to indicate that the model of the trained machine learning model obtained by the target participant node converges in the federal agency if the model prediction accuracy of the trained machine learning model obtained by the target participant node is close to or the same as the model prediction accuracy of the trained machine learning model obtained by n participant nodes before the target participant node, and the model prediction accuracy after multiple training tends to be stable, so that the training cutoff condition is considered to be satisfied. If the model prediction accuracy of the trained machine learning model obtained by the target participant node is greatly different from the model prediction accuracy of the trained machine learning model obtained by n participant nodes before the target participant node, the situation that the model is not converged in a federal institution is indicated, and the training cutoff condition is not met.
When the training cutoff condition is that the training frequency of the trained machine learning model in the federal agency is greater than or equal to the training frequency threshold, the process of executing step S103 is that the target participant node determines whether the number of times (i.e., the training frequency) that the trained machine learning model is trained by the participant node in the federal agency is greater than or equal to the training frequency threshold, if so, the training cutoff condition is considered to be satisfied, and if not, the training cutoff condition is considered not to be satisfied. For example, if the training time threshold is set to 6, the federal organization is a ring link formed by A, B, C, D nodes connected in series in sequence, the node a starts training first, after the node a obtains a trained machine learning model, because the current training time is 1 and is less than the training time threshold 6, the training cutoff condition is not satisfied, the node a transfers model parameters of the trained machine learning model to the node B, the node B uses the received model parameters to construct a local machine learning model, then the local machine learning model is trained by using local data of the node B, the machine learning model after the node B training is obtained, at this moment, the training time is 2 and is less than 6, the training cutoff condition is not satisfied, the model parameters are continuously transferred to the node C … … for training in sequence until the training time is greater than or equal to 6, that is, when the node B receives the model parameters transmitted by the node a again and trains the trained machine model parameters again, the node B can determine that the model at this time meets the training cutoff condition.
When the training cutoff condition is that the training time of the trained machine learning model in the federal agency is greater than or equal to the time threshold, the process of executing step S103 is as follows: the target participant node judges whether the time of the machine learning model after training accumulated by each participant node in turn in the federal agency (namely the training time of the machine learning model after training in the federal agency) is greater than or equal to a time threshold, and if the time is greater than or equal to the time threshold, the machine learning model after training is considered to meet the training cutoff condition.
And S104, determining the machine learning model after the target participant node is trained as a loan risk prediction model.
And after judging that the trained machine learning model meets the training cutoff condition, determining the machine learning model trained by the target participant node as a loan risk prediction model. The loan risk prediction model is a model used by the participant nodes in the federal agency to predict the overdue condition of the pending clients of the participant nodes. The target participant node can predict the overdue condition of the client to be checked by using the loan risk prediction model, and then select whether to pass the loan application of the client to be checked according to the predicted overdue condition.
Before the trained machine learning model is judged to meet the training cutoff condition, the model is alternately trained by the participant nodes in a federal organization in a series sequence until the training cutoff condition is met, the training is terminated, the trained machine learning model is trained by using credit characteristic data of local clients and actual overdue condition data of the clients of the participant nodes, unlike the prior art, only the credit characteristic data of the local clients of one participant node and the actual overdue condition data of the clients are used for training, and in the training process, only model parameters flow among the participant nodes, but the credit characteristic data of the clients of the participant nodes are not transmitted to other participant nodes by using the actual overdue condition data of the clients, so that the safety of local data and the privacy of the clients are guaranteed, and a better loan risk prediction model trained by using the sample data of a plurality of participant nodes is obtained.
Optionally, in a specific embodiment of the present application, after the step S104 is executed, the method further includes:
and sending the model parameters of the loan risk prediction model to each participant node in the federal institution, and respectively constructing the loan risk prediction model by each participant node by using the model parameters of the loan risk prediction model.
After the target participant node determines the loan risk prediction model, the loan risk prediction model can be synchronously used by each participant node in the federal organization, specifically, model parameters of the loan risk prediction model are respectively sent to each participant node in the federal organization, after the participant nodes receive the model parameters, the participant nodes use the model parameters of the loan risk prediction model to construct the loan risk prediction model, and then the loan risk prediction model is used for predicting overdue conditions of customers.
And S105, sending the model parameters of the machine learning model after the target participant node is trained to the next participant node, and determining the machine learning model meeting the training cutoff condition as the loan risk prediction model until the participant node in the alliance is trained to obtain the machine learning model meeting the training cutoff condition.
The loan risk prediction model is a model used by the participant nodes in the federal agency to predict the overdue condition of the pending clients of the participant nodes. Since the machine learning model after the target participant node training does not meet the training cutoff condition, the model needs to continue to be trained by the next participant node. Specifically, the model parameters of the machine learning model after the target participant node is trained are sent to the next participant node, the next participant node constructs a local initial machine learning model of the next participant node by using the model parameters of the machine learning model after the target participant node is trained (the local initial machine learning model of the next participant node and the machine learning model after the target participant node is trained are the same model), and then the next participant node continues to execute the embodiment shown in fig. 1 until the participant node in the alliance has a machine learning model meeting the training cutoff condition, and determines the machine learning model meeting the training cutoff condition as the loan risk prediction model. Specifically, if the other participant nodes except the target participant node obtain the machine learning model meeting the training cutoff condition and determine the machine learning model as the loan risk prediction model, the target participant node receives the model parameters of the loan risk prediction model, and the loan risk prediction model is constructed by using the model parameters of the loan risk prediction model. If the target participant node obtains the machine learning model meeting the training cutoff condition, after the target participant node is determined to be the loan risk prediction model, the model parameters of the loan risk prediction model are sent to each other participant node in the federal organization.
Optionally, referring to fig. 4, in an embodiment of the present application, after determining the machine learning model after the target participant node training as the loan risk prediction model, the method further includes:
s401, obtaining credit characteristic data of the client to be approved.
Wherein the pending clients are clients who have submitted loan applications that have not yet been approved. The credit feature data mentioned in step S401 is the same as the credit feature data mentioned in the above embodiments, and is not described herein again.
S402, inputting the credit characteristic data of the client to be approved into the loan risk prediction model, and outputting the predicted overdue condition data of the client to be approved by the loan risk prediction model.
The predicted overdue condition data of the client to be approved is used for explaining whether the loan risk prediction model predicts the overdue repayment of the client to be approved. The predicted overdue condition data of the client to be examined and approved can be a predicted probability value of overdue payment of the client to be examined and approved or a numerical value representing whether the client to be examined and approved carries out overdue payment or not. For example, if the predicted overdue condition data of the client to be checked is 1, the client to be checked represents overdue payment, and if the predicted overdue condition data of the client to be checked is 0, the client to be checked represents no overdue payment.
And S403, rejecting the loan application of the client to be approved if the predicted overdue condition data of the client to be approved indicates that the loan risk prediction model predicts the overdue repayment of the client to be approved.
If the predicted overdue condition of the pending client indicates that the loan risk prediction model predicts the overdue repayment of the pending client, the pending client is proved to have higher risk of overdue repayment, and therefore the loan application of the pending client needs to be rejected to avoid the risk.
S404, accepting the loan application of the client to be examined if the predicted overdue condition data of the client to be examined shows that the loan risk prediction model predicts that the client to be examined will not pay overdue.
If the predicted overdue condition of the client to be examined indicates that the loan risk prediction model predicts that the client to be examined does not pay overdue, the client to be examined is proved to have higher credit and lower risk of paying overdue, so that the client to be examined can accept the loan application.
The method for constructing the risk prediction model provided by the embodiment of the application is applied to a target participant node, the target participant node refers to each participant node in a federal organization, the federal organization is composed of all participant nodes which are connected in series according to a specific sequence, and the machine learning model after the target participant node is trained is obtained by obtaining credit characteristic data of each local client and actual overdue condition data of each client and then training the local initial machine learning model by using the credit characteristic data of each client and the actual overdue condition data of each client. If the target participant node is the first participant node to be trained in the federal agency, the model parameters of the local initial machine model are the initial model parameters, and if the target participant node is not the participant node of the first training model in the federal agency, the model parameters of the local initial machine model are the model parameters of the machine learning model trained by the previous participant node. And if the machine learning model after the target participant node training meets the training cutoff condition, determining the machine learning model after the target participant node training as a loan risk prediction model. And if the machine learning model after the target participant node training does not meet the training cutoff condition, sending the model parameters of the machine learning model after the target participant node training to the next participant node, and determining the machine learning model meeting the training cutoff condition as the loan risk prediction model until the participant node training in the alliance obtains the machine learning model meeting the training cutoff condition. The loan risk prediction model in the embodiment of the application is obtained by training the credit characteristic data of each client local to each participant node and the actual overdue condition data of each client, so that the used client sample data is richer than that in the prior art, and only model parameters of the machine learning model after the training of the participant nodes are transmitted among the participant nodes, but local client data is not transmitted, so that the security of local data and the privacy of the clients are guaranteed, and a high-quality model is constructed.
Referring to fig. 5, based on the above method for constructing a risk prediction model provided in the embodiment of the present application, the embodiment of the present application correspondingly discloses a device for constructing a risk prediction model, which is applied to a target participant node, where the target participant node refers to each participant node in a federal organization, and the federal organization is composed of each participant node connected in series according to a specific sequence, and the device for constructing a risk prediction model includes: a first obtaining unit 501, a training unit 502, a first determining unit 503, and a first sending unit 504.
A first obtaining unit 501, configured to obtain credit feature data of each local client and actual overdue condition data of each client. The actual overdue condition data of the client is used for explaining whether the client actually has overdue payment behaviors at the historical time point.
Optionally, in a specific embodiment of the present application, the data type of the credit feature data acquired by the target participant node is consistent with the data type of the credit feature data acquired by each participant node in the federal organization.
The training unit 502 is configured to train the local initial machine learning model by using the credit feature data of each client and the actual overdue condition data of each client, so as to obtain a machine learning model after the target participant node is trained. If the target participant node is the first participant node to be trained in the federal agency, the model parameters of the local initial machine model are the initial model parameters, and if the target participant node is not the participant node of the first training model in the federal agency, the model parameters of the local initial machine model are the model parameters of the machine learning model trained by the previous participant node.
Optionally, in an embodiment of the present application, the training unit 502 includes: a first input subunit and an adjustment subunit.
And the first input subunit is used for inputting the credit characteristic data of each client into the local initial machine learning model, and outputting the predicted overdue condition data of each client by the local initial machine learning model. The predicted overdue condition data of the client is used for explaining whether the client is overdue and paid or not predicted by the local initial machine learning model.
And the adjusting subunit is used for continuously adjusting the model parameters in the local initial machine learning model by using the error between the predicted overdue condition data of each client and the actual overdue condition data of the client until the error between the predicted overdue condition data of each client output by the adjusted local initial machine learning model and the actual overdue condition data of the client meets a preset convergence condition, and determining the adjusted local initial machine learning model as the machine learning model trained by the target participant node.
A first determining unit 503, configured to determine the machine learning model trained by the target participant node as the loan risk prediction model if the machine learning model trained by the target participant node satisfies the training cutoff condition.
Optionally, in a specific embodiment of the present application, the training cutoff condition is: the model of the trained machine learning model is converged in the federal agency, the training times of the trained machine learning model in the federal agency are larger than or equal to a threshold value of the training times, or the training time of the trained machine learning model in the federal agency is larger than or equal to a time threshold value.
The first sending unit 504 is configured to send the model parameters of the machine learning model after the target participant node is trained to the next participant node if the machine learning model after the target participant node is trained does not meet the training cutoff condition, and determine the machine learning model meeting the training cutoff condition as the loan risk prediction model until the participant node in the federation trains to obtain the machine learning model meeting the training cutoff condition. The loan risk prediction model is a model used by the participant nodes in the federal agency to predict the overdue condition of the pending clients of the participant nodes.
Optionally, in a specific embodiment of the present application, the method further includes: the device comprises a second acquisition unit, a second input subunit, a rejector subunit and a receiving subunit.
And the second acquisition unit is used for acquiring the credit characteristic data of the client to be approved.
And the second input subunit is used for inputting the credit characteristic data of the client to be approved to the loan risk prediction model and outputting the predicted overdue condition data of the client to be approved by the loan risk prediction model. The predicted overdue condition data of the client to be approved is used for explaining whether the loan risk prediction model predicts the overdue repayment of the client to be approved.
And the rejecting subunit is used for rejecting the loan application of the client to be approved if the predicted overdue condition data of the client to be approved indicates that the loan risk prediction model predicts the overdue repayment of the client to be approved.
And the acceptance subunit is used for accepting the loan application of the client to be approved if the predicted overdue condition data of the client to be approved indicates that the loan risk prediction model predicts that the client to be approved cannot pay overdue.
Optionally, in a specific embodiment of the present application, the method further includes:
and the second sending unit is used for sending the model parameters of the loan risk prediction model to each participant node in the federal institution, and building the loan risk prediction model by each participant node by using the model parameters of the loan risk prediction model.
Optionally, in an embodiment of the present application, the credit feature data of the client includes: at least one of transaction data for the customer, asset data for the customer, and blacklist record data for the customer.
The specific principle and the execution process of each unit in the device for constructing a risk prediction model disclosed in the embodiment of the present application are the same as the method for constructing a risk prediction model disclosed in the embodiment of the present application, and reference may be made to corresponding parts in the method for constructing a risk prediction model disclosed in the embodiment of the present application, which are not described herein again.
The risk prediction model construction device provided by the embodiment of the application is applied to a target participant node, the target participant node refers to each participant node in a federal organization, the federal organization is composed of each participant node which is connected in series according to a specific sequence, a first acquisition unit 501 acquires credit characteristic data of each local client and actual overdue condition data of each client, and a training unit 502 trains a local initial machine learning model by using the credit characteristic data of each client and the actual overdue condition data of each client to obtain the machine learning model after the target participant node is trained. If the target participant node is the first participant node to be trained in the federal agency, the model parameters of the local initial machine model are the initial model parameters, and if the target participant node is not the participant node of the first training model in the federal agency, the model parameters of the local initial machine model are the model parameters of the machine learning model trained by the previous participant node. If the machine learning model after the training of the target participant node satisfies the training cutoff condition, the first determining unit 503 determines the machine learning model after the training of the target participant node as the loan risk prediction model. If the machine learning model after the target participant node training does not meet the training cutoff condition, the first sending unit 504 sends the model parameters of the machine learning model after the target participant node training to the next participant node until the participant node training in the alliance obtains the machine learning model meeting the training cutoff condition, and then determines the machine learning model meeting the training cutoff condition as the loan risk prediction model. The loan risk prediction model in the embodiment of the application is obtained by training the credit characteristic data of each client local to each participant node and the actual overdue condition data of each client, so that the used client sample data is richer than that in the prior art, and only model parameters of the machine learning model after the participant nodes are trained are transmitted among the participant nodes, and the local client data is not transmitted, so that the security of local data and the privacy of the clients are guaranteed, and a high-quality model is constructed.
The embodiment of the application discloses a computer readable medium, on which a computer program is stored, wherein the program is executed by a processor to implement the method for constructing the risk prediction model according to the above embodiments.
The embodiment of the application discloses equipment, includes: one or more processors, a storage device, and one or more programs stored thereon, which when executed by the one or more processors, cause the one or more processors to implement the method for constructing a risk prediction model as described in the embodiments above.
The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
It is further noted that, herein, relational terms such as first and second, and the like may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other identical elements in a process, method, article, or apparatus that comprises the element.
Claims (10)
1. A construction method of a risk prediction model is characterized by being applied to target participant nodes, wherein the target participant nodes refer to each participant node in a federal organization; the federal agency is composed of each of the participant nodes connected in series in a specific order; the construction method of the risk prediction model comprises the following steps:
acquiring credit characteristic data of each local client and actual overdue condition data of each client; the actual overdue condition data of the client is used for explaining whether the client actually has overdue repayment behavior at a historical time point;
training a local initial machine learning model by using credit characteristic data of each client and actual overdue condition data of each client to obtain a machine learning model after the target participant node is trained; if the target participant node is the first participant node to be trained in the federal agency, the model parameters of the local initial machine model are initial model parameters; if the target participant node is not the participant node of the first training model in the federal agency, the model parameters of the local initial machine model are the model parameters of the machine learning model trained by the previous participant node;
if the machine learning model trained by the target participant node meets a training cutoff condition, determining the machine learning model trained by the target participant node as a loan risk prediction model;
if the machine learning model trained by the target participant node does not meet the training cutoff condition, sending model parameters of the machine learning model trained by the target participant node to a next participant node until the participant node in the alliance trains to obtain the machine learning model meeting the training cutoff condition, and determining the machine learning model meeting the training cutoff condition as the loan risk prediction model; the loan risk prediction model is a model used by a participant node in the federal agency to predict the overdue condition of the pending clients of the participant node.
2. The method of claim 1, wherein the training cutoff condition is: the model of the trained machine learning model is converged in the federal agency, the training times of the trained machine learning model in the federal agency are larger than or equal to a threshold value of the training times, or the training time of the trained machine learning model in the federal agency is larger than or equal to a time threshold value.
3. The method of claim 1, wherein the training a local initial machine learning model using the credit profile data of each customer and the actual overdue condition data of each customer to obtain the machine learning model trained by the target participant node comprises:
inputting the credit characteristic data of each customer into a local initial machine learning model, and outputting the predicted overdue condition data of each customer by the local initial machine learning model; the predicted overdue condition data of the client is used for explaining whether the client is overdue and repayment predicted by the local initial machine learning model;
and continuously adjusting model parameters in the local initial machine learning model by using the error between the predicted overdue condition data of each client and the actual overdue condition data of the client until the error between the predicted overdue condition data of each client and the actual overdue condition data of the client output by the adjusted local initial machine learning model meets a preset convergence condition, and determining the adjusted local initial machine learning model as the machine learning model trained by the target participant node.
4. The method of claim 1, wherein after determining the trained machine learning model of the target participant node as a loan risk prediction model, further comprising:
acquiring credit characteristic data of a client to be examined and approved;
inputting the credit characteristic data of the client to be approved into the loan risk prediction model, and outputting the predicted overdue condition data of the client to be approved by the loan risk prediction model; the predicted overdue condition data of the client to be approved is used for explaining whether the loan risk prediction model predicts the overdue repayment of the client to be approved;
if the predicted overdue condition data of the client to be checked shows that the loan risk prediction model predicts the overdue repayment of the client to be checked, rejecting the loan application of the client to be checked;
and if the predicted overdue condition data of the client to be checked shows that the loan risk prediction model predicts that the client to be checked will not be overdue for repayment, accepting the loan application of the client to be checked.
5. The method of claim 1, wherein the data type of the credit profile data obtained by the target participant node is consistent with the data type of the credit profile data obtained by each of the participant nodes in the federal agency.
6. The method of claim 1, wherein after determining the machine learning model trained by the target participant node as a loan risk prediction model if the machine learning model trained by the target participant node satisfies a training cutoff condition, further comprising:
and sending the model parameters of the loan risk prediction model to each participant node in the federal organization, and respectively constructing the loan risk prediction model by each participant node by using the model parameters of the loan risk prediction model.
7. The method of claim 1, wherein the credit profile data of the client comprises: at least one of transaction data for the customer, asset data for the customer, and blacklist record data for the customer.
8. A risk prediction model construction device is applied to target participant nodes, wherein the target participant nodes refer to each participant node in a federal organization; the federal agency is composed of each of the participant nodes connected in series in a specific order; the risk prediction model construction device comprises:
the system comprises a first acquisition unit, a second acquisition unit and a third acquisition unit, wherein the first acquisition unit is used for acquiring credit characteristic data of each local client and actual overdue condition data of each client; the actual overdue condition data of the client is used for explaining whether the client actually has overdue repayment behavior at a historical time point;
the training unit is used for training a local initial machine learning model by using the credit characteristic data of each client and the actual overdue condition data of each client to obtain the machine learning model after the node of the target participant is trained; if the target participant node is the first participant node to be trained in the federal agency, the model parameters of the local initial machine model are initial model parameters; if the target participant node is not the participant node of the first training model in the federal agency, the model parameters of the local initial machine model are the model parameters of the machine learning model trained by the previous participant node;
the first determining unit is used for determining the machine learning model trained by the target participant node as a loan risk prediction model if the machine learning model trained by the target participant node meets a training cutoff condition;
a first sending unit, configured to send, if the machine learning model trained by the target participant node does not meet the training cutoff condition, a model parameter of the machine learning model trained by the target participant node to a next participant node, and determine, until there is a machine learning model meeting the training cutoff condition in the coalition enterprises, the machine learning model meeting the training cutoff condition as the loan risk prediction model when the participant node is trained to obtain the machine learning model meeting the training cutoff condition; the loan risk prediction model is a model used by a participant node in the federal agency to predict the overdue condition of the pending clients of the participant node.
9. A computer-readable medium, on which a computer program is stored, wherein the program, when executed by a processor, implements the method of any one of claims 1 to 7.
10. An apparatus, comprising:
one or more processors;
a storage device having one or more programs stored thereon;
the one or more programs, when executed by the one or more processors, cause the one or more processors to implement the method of any of claims 1-7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110661850.6A CN113240512A (en) | 2021-06-15 | 2021-06-15 | Method, device, readable medium and equipment for constructing risk prediction model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110661850.6A CN113240512A (en) | 2021-06-15 | 2021-06-15 | Method, device, readable medium and equipment for constructing risk prediction model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113240512A true CN113240512A (en) | 2021-08-10 |
Family
ID=77139971
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110661850.6A Pending CN113240512A (en) | 2021-06-15 | 2021-06-15 | Method, device, readable medium and equipment for constructing risk prediction model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113240512A (en) |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108846520A (en) * | 2018-06-22 | 2018-11-20 | 北京京东金融科技控股有限公司 | Overdue loan prediction technique, device and computer readable storage medium |
| CN111310938A (en) * | 2020-02-10 | 2020-06-19 | 深圳前海微众银行股份有限公司 | Semi-supervision-based horizontal federal learning optimization method, equipment and storage medium |
| CN111461874A (en) * | 2020-04-13 | 2020-07-28 | 浙江大学 | Credit risk control system and method based on federal mode |
| CN111553484A (en) * | 2020-04-30 | 2020-08-18 | 同盾控股有限公司 | Method, device and system for federal learning |
| CN111967910A (en) * | 2020-08-18 | 2020-11-20 | 中国银行股份有限公司 | User passenger group classification method and device |
| CN112270597A (en) * | 2020-11-10 | 2021-01-26 | 恒安嘉新（北京）科技股份公司 | Business processing and credit evaluation model training method, device, equipment and medium |
| CN112906903A (en) * | 2021-01-11 | 2021-06-04 | 北京源堡科技有限公司 | Network security risk prediction method and device, storage medium and computer equipment |
-
2021
- 2021-06-15 CN CN202110661850.6A patent/CN113240512A/en active Pending
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108846520A (en) * | 2018-06-22 | 2018-11-20 | 北京京东金融科技控股有限公司 | Overdue loan prediction technique, device and computer readable storage medium |
| CN111310938A (en) * | 2020-02-10 | 2020-06-19 | 深圳前海微众银行股份有限公司 | Semi-supervision-based horizontal federal learning optimization method, equipment and storage medium |
| CN111461874A (en) * | 2020-04-13 | 2020-07-28 | 浙江大学 | Credit risk control system and method based on federal mode |
| CN111553484A (en) * | 2020-04-30 | 2020-08-18 | 同盾控股有限公司 | Method, device and system for federal learning |
| CN111967910A (en) * | 2020-08-18 | 2020-11-20 | 中国银行股份有限公司 | User passenger group classification method and device |
| CN112270597A (en) * | 2020-11-10 | 2021-01-26 | 恒安嘉新（北京）科技股份公司 | Business processing and credit evaluation model training method, device, equipment and medium |
| CN112906903A (en) * | 2021-01-11 | 2021-06-04 | 北京源堡科技有限公司 | Network security risk prediction method and device, storage medium and computer equipment |
Non-Patent Citations (1)
| Title |
|---|
| LAVENDER_LOVEQAQ: "《联邦学习》-个人笔记（四）", 《CSDN》HTTPS://BLOG.CSDN.NET/WEIXIN_42394170/ARTICLE/DETAILS/109466352, 3 November 2020 (2020-11-03) * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110147925B (en) | Risk decision method, device, equipment and system | |
| US20190392314A1 (en) | Method and device for cash advance recognition | |
| CN111428945A (en) | Business audit processing method and device and electronic equipment | |
| CN111091460B (en) | Data processing method and device | |
| US11334878B2 (en) | Combining explicit and implicit feedback in self-learning fraud detection systems | |
| US20230262160A1 (en) | Systems and methods for phishing monitoring | |
| CN110705988A (en) | Interactive blocking method and system for invasive behaviors | |
| CN112819187A (en) | Management method and system of shared instrument | |
| CN115099825A (en) | Abnormal fund flow account identification method and device | |
| WO2022204242A1 (en) | Automated claims adjustment assignment utilizing crowdsourcing and adjuster priority score determinations | |
| CN113240512A (en) | Method, device, readable medium and equipment for constructing risk prediction model | |
| CN112053245A (en) | Information evaluation method and system | |
| US20240037585A1 (en) | Using machine learning model to automatically predict updated assessment score | |
| US11210638B2 (en) | Systems and methods for providing contextual calendar reminders | |
| US20240037406A1 (en) | Testing predicted data utilizing trained machine learning model | |
| CN110324418A (en) | Method and apparatus based on customer relationship transmission service | |
| US12026718B1 (en) | Low latency anomaly detection and related mitigation action | |
| CN115330490A (en) | Product recommendation method and device, storage medium and equipment | |
| US20230012458A1 (en) | Identifying transaction processing retry attempts based on machine learning models for transaction success | |
| CN115150289A (en) | Exception handling method and system based on composite monitoring | |
| Phoomvuthisarn | A survey study on reputation-based trust mechanisms in service-oriented computing | |
| US20200394067A1 (en) | Systems and methods for pre-executing idiosyncratic computation through the application of algorithmic prediction of user behavior patterns | |
| JP2003030441A (en) | System and method for adjustment | |
| WO2020174483A1 (en) | Prediction of billing disputes | |
| Shakeri et al. | A layer model of a confidence-aware trust management system |
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 |