CN111882426A - Business risk classifier training method, device, equipment and storage medium - Google Patents
Business risk classifier training method, device, equipment and storage medium Download PDFInfo
- Publication number
- CN111882426A CN111882426A CN202010703942.1A CN202010703942A CN111882426A CN 111882426 A CN111882426 A CN 111882426A CN 202010703942 A CN202010703942 A CN 202010703942A CN 111882426 A CN111882426 A CN 111882426A
- Authority
- CN
- China
- Prior art keywords
- risk
- subtasks
- training
- business
- sub
- 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.)
- Granted
Links
- 238000012549 training Methods 0.000 title claims abstract description 121
- 238000000034 method Methods 0.000 title claims abstract description 34
- 238000003860 storage Methods 0.000 title claims abstract description 16
- 238000012360 testing method Methods 0.000 claims abstract description 67
- 238000005070 sampling Methods 0.000 claims description 20
- 238000004590 computer program Methods 0.000 claims description 17
- 230000002596 correlated effect Effects 0.000 claims description 12
- 230000000875 corresponding effect Effects 0.000 claims description 12
- 238000010586 diagram Methods 0.000 description 11
- 230000008569 process Effects 0.000 description 10
- 238000010801 machine learning Methods 0.000 description 8
- 230000006870 function Effects 0.000 description 6
- 238000012545 processing Methods 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 3
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000007635 classification algorithm Methods 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 230000005055 memory storage Effects 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 238000007637 random forest analysis Methods 0.000 description 1
- 238000000638 solvent extraction Methods 0.000 description 1
- 230000003068 static effect Effects 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Business, Economics & Management (AREA)
- Software Systems (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Accounting & Taxation (AREA)
- Finance (AREA)
- Bioinformatics & Computational Biology (AREA)
- Development Economics (AREA)
- Marketing (AREA)
- Strategic Management (AREA)
- Technology Law (AREA)
- General Business, Economics & Management (AREA)
- Medical Informatics (AREA)
- Economics (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The present specification provides a business risk classifier training method, apparatus, device and storage medium, the method comprising: dividing the service data into a first training set, a first test set, a second training set and a second test set; generating a plurality of first subtasks using a first training set; the same negative sample is marked differently in different first subtasks, and the positive and negative samples in each first subtask are in a specified proportion; performing meta-training on the first learner by using the first subtask and the first test set to obtain an optimal learning parameter; generating a plurality of second subtasks by using a second training set, respectively inputting the second subtasks into a first learner with the optimal learning parameters as parameters, and generating a plurality of sub-risk classifiers according to corresponding output; testing the plurality of sub-risk classifiers by using a second test set to obtain a plurality of predictor results; and combining the predictor results to obtain a risk prediction result. The method and the system can improve the recall rate and the prediction accuracy rate of the business risk classifier under the condition of not forming a negative case.
Description
Technical Field
The present disclosure relates to the field of machine learning technologies, and in particular, to a method, an apparatus, a device, and a storage medium for training a business risk classifier.
Background
Some current business risk classifiers (e.g., loan overdue risk prediction for banking loan businesses, etc.) are based on machine learning. Generally, the number of positive samples in the service data far exceeds the number of negative samples (that is, the proportion of the positive samples to the negative samples is unbalanced), and when the machine learning training is performed on the initial model in this scenario, the training of the negative samples is easily insufficient, so that the recall rate (recall) of the trained service risk classifier is easily low.
In order to solve the problem of the unbalanced proportion of the positive samples and the negative samples, the number of the negative samples is generally increased by adopting a mode of fictitious negative sample data in the prior art. However, the negative examples generated by the fiction are not interpretable and have no practical significance. Moreover, the way of making up the data does not comply with the rules of the financial security regulations. Therefore, the business risk classifier obtained by machine learning training using fictive negative case data is difficult to be applied to risk classification and prediction scenes in the financial industry. Therefore, how to improve the recall rate of the business risk classifier on the premise of not forming negative case data is a technical problem to be solved urgently at present.
Disclosure of Invention
An object of an embodiment of the present specification is to provide a method, an apparatus, a device, and a storage medium for training a business risk classifier, so as to improve a recall rate of the business risk classifier on the premise of not forming negative case data.
In order to achieve the above object, in one aspect, an embodiment of the present specification provides a business risk classifier training method, including:
dividing the acquired business data into a first training set and a first test set for meta-training, and a second training set and a second test set for meta-testing;
sampling a plurality of first subtasks from the first training set; the marks of the same negative sample in different first subtasks are different, and the positive sample and the negative sample in each first subtask are in a specified ratio;
performing meta-training on a first learner by using the plurality of first subtasks and the first test set to obtain an optimal learning parameter;
sampling a plurality of second subtasks from the second training set, respectively inputting the plurality of second subtasks into a first learner with the optimal learning parameters as parameters, and generating a plurality of sub-risk classifiers according to corresponding outputs;
testing the plurality of sub-risk classifiers by using the second test set to correspondingly obtain a plurality of sub-risk prediction results;
and combining the multiple risk predictor results to obtain a risk prediction result.
In one embodiment of the present disclosure, the same negative sample is labeled differently in different second subtasks, and the positive and negative samples in each second subtask have a specified ratio.
In one embodiment of the present specification, the data amount of each class of business risk category in each of the first subtasks is positively correlated with the total number of negative samples in the first training set; the data volume of each type of business risk category in each second subtask is positively correlated with the total number of negative samples in the second training set.
In an embodiment of the present disclosure, the combining the plurality of risk predictors to obtain a risk prediction result comprises:
and performing majority voting on the plurality of risk prediction sub-results, and determining a risk prediction result according to the voting result.
In another aspect, an embodiment of the present specification further provides a business risk classifier training device, including:
the data dividing module is used for dividing the acquired business data into a first training set and a first test set for meta-training, and a second training set and a second test set for meta-testing;
a task generation module for sampling a plurality of first subtasks from the first training set; the marks of the same negative sample in different first subtasks are different, and the positive sample and the negative sample in each first subtask are in a specified ratio;
the parameter acquisition module is used for performing meta-training on a first learner by utilizing the plurality of first subtasks and the first test set to obtain an optimal learning parameter;
the model generation module is used for sampling a plurality of second subtasks from the second training set, inputting the plurality of second subtasks into a first learner with the optimal learning parameter as a parameter respectively, and generating a plurality of sub-risk classifiers according to corresponding output;
the model testing module is used for testing the sub-risk classifiers by utilizing the second testing set to correspondingly obtain a plurality of sub-risk prediction results;
and the result combination module is used for combining the plurality of risk predictor results to obtain a risk prediction result.
In one embodiment of the present disclosure, the same negative sample is labeled differently in different second subtasks, and the positive and negative samples in each second subtask have a specified ratio.
In one embodiment of the present specification, the data amount of each class of business risk category in each of the first subtasks is positively correlated with the total number of negative samples in the first training set; the data volume of each type of business risk category in each second subtask is positively correlated with the total number of negative samples in the second training set.
In an embodiment of the present disclosure, the combining the plurality of risk predictors to obtain a risk prediction result comprises:
and performing majority voting on the plurality of risk prediction sub-results, and determining a risk prediction result according to the voting result.
In another aspect, the present specification further provides an electronic device, which includes a memory, a processor, and a computer program stored on the memory, where the computer program is executed by the processor to perform the business risk classifier training method described above.
In another aspect, the present specification further provides a storage medium, on which a computer program is stored, and the computer program, when executed by a processor, implements the business risk classifier training method described above.
It can be seen from the above technical solutions provided by the embodiments of the present specification that, according to the embodiments of the present specification, different labels can be marked for the same negative sample in different subtasks, and the positive and negative samples in each subtask have a specified ratio, so that the same negative sample can appear in different data forms in different subtasks, thereby realizing multiplexing of the negative sample among different subtasks, that is, realizing that the ratio of the positive and negative samples in each subtask is relatively balanced under the condition of not falsely constructing the negative sample, thereby solving the problem of unbalanced positive and negative samples in the service data, and thus improving the recall rate of the service risk classifier. Moreover, the embodiments of the present disclosure may also combine the multiple risk prediction sub-results to obtain a risk prediction result, thereby facilitating to improve the prediction accuracy of the business risk classifier.
Drawings
In order to more clearly illustrate the embodiments of the present specification 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 some embodiments described in the specification, and other drawings can be obtained by those skilled in the art without inventive labor. In the drawings:
FIG. 1 is a flow chart of a meta-learning based business risk classifier training method in an embodiment provided herein;
FIG. 2 is a schematic diagram of data set partitioning in an embodiment provided herein;
FIG. 3 is a schematic illustration of meta-training in an embodiment provided herein;
FIG. 4 is a schematic illustration of meta-testing in embodiments provided herein;
FIG. 5 is a block diagram of an electronic device in an embodiment provided herein;
fig. 6 is a block diagram of a meta-learning based business risk classifier training device in an embodiment provided in the present specification.
Detailed Description
In order to make the technical solutions in the present specification better understood, the technical solutions in the embodiments of the present specification will be clearly and completely described below with reference to the drawings in the embodiments of the present specification, and it is apparent that the described embodiments are only a part of the embodiments of the present specification, but not all of the embodiments. All other embodiments obtained by a person skilled in the art without making creative efforts based on the embodiments in the present specification shall fall within the protection scope of the present specification.
The meta-learning is provided for solving the learning problem of the machine of the small sample, namely, because the total amount of data of the target field is less and does not support training prediction, another field with enough data amount can be selected for training to obtain the learning ability, and then the obtained learning ability is migrated and applied to the target field. In the embodiment of the present specification, meta learning is migrated to a data classification scenario in which the total amount of target domain data is sufficient but the proportion of positive samples and negative samples is unbalanced, so as to solve the technical problem that how to increase the recall rate of the business risk classifier is urgently needed to be solved on the premise of not forming negative sample data. The following is a detailed description of embodiments in connection with the present specification.
Referring to fig. 1, a meta-learning based business risk classifier training method according to some embodiments of the present description may include the following steps:
s101, dividing the acquired business data into a first training set and a first test set for meta-training, and a second training set and a second test set for meta-testing.
In the embodiments of the present specification, the acquired service data may depend on a specific application scenario. For example, in an exemplary embodiment, the business data may include basic information of the user, property information, liability information, historical repayment records, historical overdue records, historical earning records, and people's credit information, taking the loan overdue risk prediction of the bank loan business as an example.
In one embodiment of the present description, after the business data is obtained, it may be first preprocessed to process it into structured data that can be processed by a machine learning model and stored in a unified, standard format, thereby forming a data set.
In an embodiment of the present specification, before sampling, the preprocessed traffic data may be divided into a training set and a test set according to a preset ratio. The samples may be hierarchical random samples (or samples in other manners), and the following description of the samples related to the subtasks may refer to this portion, which is not described in detail herein.
For example, in an exemplary embodiment, the training set and test set may be divided by a ratio of 80%, 20% (e.g., as shown in FIG. 2). Those skilled in the art will appreciate that in other embodiments, the predetermined proportion may be divided according to actual needs, but generally care should be taken to maintain the consistency of the data distribution so as to avoid introducing additional bias into the data dividing process to affect the final result. In order to adapt to meta-learning, the divided training set and test set need to be further divided. Specifically, the training set may be further divided into a first training set and a first test set according to a preset ratio (e.g., 80%, 20% shown in fig. 2) for meta training (meta train). The test set may be further divided into a second training set and a second test set according to a preset ratio (e.g., 80%, 20% as shown in fig. 2) for meta test (meta test).
S102, sampling a plurality of first subtasks from the first training set; the marks of the same negative sample in different first subtasks can be different, and the positive and negative samples in each first subtask have a specified ratio.
Since meta-learning uses Task (Task) as its training data, and the training Set (also called support Set) and the test Set (also called Query Set) of meta-learning are composed of subtasks. Therefore, in order to perform meta-training, the first training set needs to be further divided (i.e., sample division) to generate a plurality of K-way-N-shot first subtasks. Wherein K refers to the number of business risk categories included in each first subtask, and N refers to the number of samples to be extracted for each category. K may take a value according to an application scenario, for example, taking two classified business risk categories (i.e., no business risk and business risk) as an example, K equals 2. In order to make the classifier in meta-learning approach the effect of the ordinary machine learning classifier, the value of N can be positively correlated with the total number of negative samples in the first training set through data experiment exploration. For example, in an exemplary embodiment, the value of N may be a specified percentage of the total number of negative examples in the first training set.
In the embodiment of the present specification, in view of the problem that positive and negative samples in the service data are unbalanced (i.e., there are fewer negative samples), different labels may be marked for the same negative sample in different first subtasks, and the positive and negative samples in each first subtask have a specified ratio (e.g., 1:1, 2:1, etc.), so that the same negative sample can appear in different data forms in different first subtasks, thereby implementing multiplexing of the negative sample among different first subtasks, that is, implementing that the ratio of the positive and negative samples in each first subtask is relatively balanced without imaginary negative samples, thereby solving the problem of unbalanced positive and negative samples in the service data.
The implementation scheme of the specification can be applied to various classification and prediction scenes which do not require to generate false data in the banking industry, and has the universality and practicability of application. In fact, the technical solution for solving the problem of imbalance between positive and negative samples in the service data in the embodiment of the present specification can be applied to most of the conventional machine learning classification algorithms, and has a wider application range, higher extensibility, and higher flexibility.
In some embodiments of the present description, when each sub-task is obtained using hierarchical random sampling, its sampling error is smaller when the number of samples is the same, compared to simple random sampling and equidistant random sampling; when the sampling error requirement is the same, it requires a smaller number of samples.
S103, meta-training the first learner by utilizing the plurality of first subtasks and the first test set to obtain the optimal learning parameters.
For ease of understanding, the meta training process in meta learning is explained below. Generally, the meta-training process mainly includes the following steps:
(1) and randomly initializing the learning parameters of the first learner, inputting a first subtask into the first learner so as to obtain an output, and generating the second learner by taking the output as the parameters.
For example, taking the exemplary embodiment shown in FIG. 3 as an example, if the initial first learner A is represented as y1＝k1x + B, the initial second learner B is denoted y2＝k2x, the data in the first subtask is x, then k1And b is the learning parameter theta, k randomly initialized by the first learner A2Parameters randomly initialized for the second learner B. When a first subtask is input to y1＝k1In x + b, y can be obtained1An output value of phi, k2Is assigned a value of phi, whereupon the second learner B becomes y2＝φx。
(2) Inputting the first test set into the second learner to obtain a predicted output, taking the cross entropy of the predicted output and the real label as a loss, and updating the learning parameter (e.g., the learning parameter θ in fig. 3) in a reverse direction by using a mini-batch random gradient descent as a gradient descent strategy (which is only an example here, and in other embodiments, any other suitable gradient descent strategy may be selected).
(3) On the basis of the reverse updating learning parameters, inputting the next first subtask into the first learner after the reverse updating learning parameters to continue training so as to continue learning and reversely updateAnd (5) learning parameters. By repeating the above steps, after many times of iterative training learning, an optimal learning parameter (e.g., the optimal learning parameter θ corresponding to the learning parameter θ in fig. 3) can be finally obtained*). The optimal learning parameters are learning ability learned through meta-training.
It follows that in the meta-training process of embodiments of the present specification, the output of the first learner with θ as a learning parameter can be used to construct the second learner, i.e., the parameters of the second learner are determined from the output of the first learner; meanwhile, the output of the second learner may be used to optimize the learning parameter θ of the first learner, based on a preset gradient descent strategy. The main purpose of meta-training is to obtain the optimal learning parameter θ of the first learner*For subsequent meta-testing.
S104, sampling a plurality of second subtasks from the second training set, respectively inputting the second subtasks into a first learner with the optimal learning parameter as a parameter, and generating a plurality of sub-risk classifiers according to corresponding outputs.
In an embodiment of the present specification, the sampling of the plurality of second subtasks from the second training set may be performed by hierarchically and randomly sampling the second training set into a plurality of K-way-N-shot second subtasks. Similar to the first subtask, the same negative sample can be marked differently in different second subtasks, and the positive and negative samples in each second subtask have a specified ratio, so that the ratio of the positive and negative samples in each second subtask can be relatively balanced without forming a negative sample. And each second subtask includes two business risk categories, and each business risk category has a specified number of samples, and the specified number is positively correlated with the total number of negative samples in the second training set. Furthermore, to facilitate meta-testing, the amount of data in the second sub-task may be equal to the amount of data in the second sub-task.
As shown in fig. 4, since the learning ability (i.e., the above-mentioned optimal learning parameter θ) is learned already at the time of meta-training*). On the basis, the plurality of second subtasks are respectively processedInputting the optimal learning parameter theta*In the learning device A as parameter, a plurality of outputs phi can be obtained correspondingly*(e.g., as in FIG. 4)
And
etc.) at said plurality of outputs phi*A plurality of sub-risk classifiers may be generated as parameters. Therefore, learning capacity obtained by meta-training is endowed to the sub-risk classifiers, and the sub-risk classifiers do not need to be obtained from random initialization training. In FIG. 4, the input is at θ*The second subtask in learner A, being a parameter, is not the same, and results in an output φ*Such as in fig. 4
And
etc.) there will be differences so that a number of different sub-risk classifiers can be derived in the end.
In the embodiments of the present specification, the first learner and the second learner used for training may be any suitable machine learning model, and may be specifically selected by the user according to actual needs. For example, in an exemplary embodiment, the second learner may select to be a random forest classifier.
And S105, testing the plurality of sub-risk classifiers by using the second test set to correspondingly obtain a plurality of sub-risk prediction results.
In order to test the performance of each sub-risk classifier, the same batch of data of the second test set may be input into each sub-risk classifier, so that the risk prediction sub-result of each sub-risk classifier may be obtained. In one embodiment of the present description, to improve the testing efficiency, the testing may be initiated by multiple threads running in parallel.
And S106, combining the multiple risk predictor results to obtain a risk prediction result.
As can be seen from step S105, each of the sub-risk classifiers obtained above can be used for business risk prediction separately. But because the data volume of the second subtask is small during training, the problem of insufficient training of a single risk sub-classifier is easy to occur. Therefore, in order to improve the prediction accuracy of the business risk classifier, the multiple risk prediction sub-results may be combined to obtain a risk prediction result, that is, the multiple sub-risk classifiers may be used for combined prediction. In some embodiments of the present description, the combining the plurality of risk predictor results may be, for example, majority voting the plurality of risk predictor results and determining a risk predictor result according to the voting results.
In one embodiment of the present disclosure, the Majority vote may be a hard vote (Majority/Hardvoting). The basic idea is to select the class of the algorithm that outputs the most. For example, in one exemplary embodiment, there are four sub-risk classifiers c1、c2、c3And c4For the same input, the corresponding outputs are: y, Y, N and Y; wherein Y represents immediate risk and N represents no risk. Due to c1、c2And c4The risk predictor result Y of (a) accounts for the most of the four risk predictor results; thus, four sub-risk classifiers c1、c2、c3And c4The combined risk prediction results are at risk.
In one embodiment of the present disclosure, the majority vote may also be a Soft vote (Soft voting). The soft voting is also called weighted average probability voting, and is a voting method using output class probability classification, and the basic idea is to obtain a weighted average value of each class probability by configuring voting weight for each sub-risk classifier, and select the class with a larger value. Obviously, the voting weights of the sub-risk classifiers are the same in hard voting (i.e., the voting weights are all 1). However, in soft voting, the voting weights of the sub-risk classifiers may be different, which may affect the voting combination. Wherein each sub-windThe voting weight of the risk classifier may be specifically determined according to a performance index (which may include, but is not limited to, precision, recall, and/or accuracy, for example) of each sub-risk classifier. For example, when the performance index of the sub-risk classifier is high, a proper high weight can be set for the sub-risk classifier; when the performance index of the sub-risk classifier is low, it is given a suitably low weight. For example, in one exemplary embodiment, there are four sub-risk classifiers c1、c2、c3And c4For the same input, the corresponding outputs are: y, Y, N and N; wherein Y represents immediate risk and N represents no risk. If four sub-risk classifiers c1、c2、c3And c4The voting weights of (a) are 0.6, 0.7, 1, 0.9, respectively. Then after the votes are combined: the vote for Y is: (0.6Y +0.7Y)/2 ═ 0.75Y; the ticket obtaining of N is as follows: (1N +0.9N)/2 ═ 0.95N.
Obviously, the votes obtained by N are higher, so that the risk prediction result obtained by voting combination is risk-free. Compared with hard voting, soft voting has better generalization performance and overfitting resistance, thereby being beneficial to further improving the prediction accuracy of the business risk classifier.
Corresponding to the business risk classifier training method based on meta-learning, the specification further provides electronic equipment. Referring to fig. 5, in some embodiments of the present description, the electronic device may include a memory, a processor, and a computer program stored on the memory, the computer program when executed by the processor may perform the steps of:
dividing the acquired business data into a first training set and a first test set for meta-training, and a second training set and a second test set for meta-testing;
sampling a plurality of first subtasks from the first training set; the marks of the same negative sample in different first subtasks are different, and the positive sample and the negative sample in each first subtask are in a specified ratio;
performing meta-training on a first learner by using the plurality of first subtasks and the first test set to obtain an optimal learning parameter;
sampling a plurality of second subtasks from the second training set, respectively inputting the plurality of second subtasks into a first learner with the optimal learning parameters as parameters, and generating a plurality of sub-risk classifiers according to corresponding outputs;
testing the plurality of sub-risk classifiers by using the second test set to correspondingly obtain a plurality of sub-risk prediction results;
and combining the multiple risk predictor results to obtain a risk prediction result.
While the process flows described above include operations that occur in a particular order, it should be appreciated that the processes may include more or less operations that are performed sequentially or in parallel (e.g., using parallel processors or a multi-threaded environment).
Corresponding to the business risk classifier training method based on meta-learning, the specification further provides a business risk classifier training device based on meta-learning. Referring to fig. 6, in some embodiments of the present specification, the meta-learning based business risk classifier training device may include:
the data dividing module 61 may be configured to divide the acquired service data into a first training set and a first test set for meta-training, and a second training set and a second test set for meta-testing.
A task generation module 62, operable to sample a plurality of first subtasks from the first training set; the marks of the same negative sample in different first subtasks are different, and the positive sample and the negative sample in each first subtask have a specified ratio.
The parameter obtaining module 63 may be configured to perform meta-training on the first learner by using the plurality of first subtasks and the first test set, so as to obtain an optimal learning parameter.
The model generating module 64 may be configured to sample a plurality of second subtasks from the second training set, input the plurality of second subtasks into the first learner with the optimal learning parameter as a parameter, and generate a plurality of sub risk classifiers according to corresponding outputs.
The model testing module 65 may be configured to test the sub-risk classifiers by using the second test set, so as to obtain a plurality of sub-risk prediction results.
And the result combination module 66 may be configured to combine the plurality of risk predictor results to obtain a risk prediction result.
In some embodiments of the meta-learning based business risk classifier training apparatus of the present disclosure, the labels of the same negative sample in different second subtasks may be different, and the positive and negative samples in each second subtask have a specific ratio.
In the business risk classifier training device based on meta-learning according to some embodiments of the present specification, the data amount of each class of business risk category in each of the first subtasks may be positively correlated with the total number of negative samples in the first training set; the data volume of each class of business risk category in each of the second subtasks may be positively correlated with the total number of negative examples in the second training set.
In the meta-learning based business risk classifier training device according to some embodiments of the present specification, the combining the multiple risk predictor results to obtain the risk prediction result may include:
and performing majority voting on the plurality of risk prediction sub-results, and determining a risk prediction result according to the voting result.
For convenience of description, the above devices are described as being divided into various units by function, and are described separately. Of course, the functions of the various elements may be implemented in the same one or more software and/or hardware implementations of the present description.
The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the specification. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
In a typical configuration, a computing device includes one or more processors (CPUs), input/output interfaces, network interfaces, and memory.
The memory may include forms of volatile memory in a computer readable medium, Random Access Memory (RAM) and/or non-volatile memory, such as Read Only Memory (ROM) or flash memory (flash RAM). Memory is an example of a computer-readable medium.
Computer-readable media, including both non-transitory and non-transitory, removable and non-removable media, may implement information storage by any method or technology. The information may be computer readable instructions, data structures, modules of a program, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), other types of Random Access Memory (RAM), Read Only Memory (ROM), Electrically Erasable Programmable Read Only Memory (EEPROM), flash memory or other memory technology, compact disc read only memory (CD-ROM), Digital Versatile Discs (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information that can be accessed by a computing device. As defined herein, a computer readable medium does not include a transitory computer readable medium such as a modulated data signal and a carrier wave.
As will be appreciated by one skilled in the art, embodiments of the present description may be provided as a method, system, or computer program product. Accordingly, embodiments of the present description may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, embodiments of the present description may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
Embodiments of the present description may be described in the general context of computer-executable instructions, such as program modules, being executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform particular tasks or implement particular abstract data types. The described embodiments may also be practiced in distributed computing environments where tasks are performed by remote processing devices that are linked through a communications network. In a distributed computing environment, program modules may be located in both local and remote computer storage media including memory storage devices.
The embodiments in the present specification are described in a progressive manner, and the same and similar parts among the embodiments are referred to each other, and each embodiment focuses on the differences from the other embodiments. In particular, for system embodiments, because they are substantially similar to process embodiments, the description is relatively simple, and reference may be made to some descriptions of process embodiments for related points. In the description of the specification, reference to the description of the term "one embodiment", "some embodiments", "an example", "a specific example", or "some examples", etc., means that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the embodiments of the specification. In this specification, the schematic representations of the terms used above do not necessarily refer to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples. Furthermore, the various embodiments or examples and features of the various embodiments or examples described in this specification can be combined and combined by those skilled in the art without contradiction.
The above description is only an embodiment of the present application and is not intended to limit the present application. Various modifications and changes may occur to those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of the claims of the present application.
Claims (10)
1. A business risk classifier training method is characterized by comprising the following steps:
dividing the acquired business data into a first training set and a first test set for meta-training, and a second training set and a second test set for meta-testing;
sampling a plurality of first subtasks from the first training set; the marks of the same negative sample in different first subtasks are different, and the positive sample and the negative sample in each first subtask are in a specified ratio;
performing meta-training on a first learner by using the plurality of first subtasks and the first test set to obtain an optimal learning parameter;
sampling a plurality of second subtasks from the second training set, respectively inputting the plurality of second subtasks into a first learner with the optimal learning parameters as parameters, and generating a plurality of sub-risk classifiers according to corresponding outputs;
testing the plurality of sub-risk classifiers by using the second test set to correspondingly obtain a plurality of sub-risk prediction results;
and combining the multiple risk predictor results to obtain a risk prediction result.
2. The business risk classifier training method of claim 1, wherein the same negative sample is labeled differently in different second subtasks, and the positive and negative samples in each second subtask have a specified ratio.
3. The business risk classifier training method according to claim 1 or 2, wherein the data amount of each business risk category in each of the first subtasks is positively correlated with the total number of negative samples in the first training set; the data volume of each type of business risk category in each second subtask is positively correlated with the total number of negative samples in the second training set.
4. The business risk classifier training method of claim 1, wherein the combining the plurality of risk predictor results to obtain a risk prediction result comprises:
and performing majority voting on the plurality of risk prediction sub-results, and determining a risk prediction result according to the voting result.
5. A business risk classifier training device, comprising:
the data dividing module is used for dividing the acquired business data into a first training set and a first test set for meta-training, and a second training set and a second test set for meta-testing;
a task generation module for sampling a plurality of first subtasks from the first training set; the marks of the same negative sample in different first subtasks are different, and the positive sample and the negative sample in each first subtask are in a specified ratio;
the parameter acquisition module is used for performing meta-training on a first learner by utilizing the plurality of first subtasks and the first test set to obtain an optimal learning parameter;
the model generation module is used for sampling a plurality of second subtasks from the second training set, inputting the plurality of second subtasks into a first learner with the optimal learning parameter as a parameter respectively, and generating a plurality of sub-risk classifiers according to corresponding output;
the model testing module is used for testing the sub-risk classifiers by utilizing the second testing set to correspondingly obtain a plurality of sub-risk prediction results;
and the result combination module is used for combining the plurality of risk predictor results to obtain a risk prediction result.
6. The business risk classifier training device of claim 5, wherein the same negative example is labeled differently in different second subtasks, and the positive and negative examples in each second subtask are in a specified ratio.
7. The business risk classifier training device according to claim 5 or 6, wherein the data amount of each business risk category in each of the first subtasks is positively correlated with the total number of negative samples in the first training set; the data volume of each type of business risk category in each second subtask is positively correlated with the total number of negative samples in the second training set.
8. The business risk classifier training device of claim 5, wherein the combining the plurality of risk predictor results to obtain a risk prediction result comprises:
and performing majority voting on the plurality of risk prediction sub-results, and determining a risk prediction result according to the voting result.
9. An electronic device comprising a memory, a processor, and a computer program stored on the memory, wherein the computer program, when executed by the processor, performs the business risk classifier training method of any one of claims 1-4.
10. A storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the business risk classifier training method of any one of claims 1-4.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010703942.1A CN111882426B (en) | 2020-07-21 | 2020-07-21 | Method, device, equipment and storage medium for training business risk classifier |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010703942.1A CN111882426B (en) | 2020-07-21 | 2020-07-21 | Method, device, equipment and storage medium for training business risk classifier |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111882426A true CN111882426A (en) | 2020-11-03 |
| CN111882426B CN111882426B (en) | 2024-02-06 |
Family
ID=73155704
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010703942.1A Active CN111882426B (en) | 2020-07-21 | 2020-07-21 | Method, device, equipment and storage medium for training business risk classifier |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111882426B (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112767167A (en) * | 2021-02-07 | 2021-05-07 | 中国工商银行股份有限公司 | Investment transaction risk trend prediction method and device based on ensemble learning |
| CN112836742A (en) * | 2021-02-02 | 2021-05-25 | 中国工商银行股份有限公司 | System resource adjusting method, device and equipment |
| CN112836750A (en) * | 2021-02-03 | 2021-05-25 | 中国工商银行股份有限公司 | System resource allocation method, device and equipment |
| CN113299346A (en) * | 2021-04-01 | 2021-08-24 | 腾讯科技（深圳）有限公司 | Classification model training and classifying method and device, computer equipment and storage medium |
| CN113592160A (en) * | 2021-07-13 | 2021-11-02 | 北京淇瑀信息科技有限公司 | Fusion processing-based loss of contact risk prediction method and system for user equipment and computer equipment |
| CN112508698B (en) * | 2021-02-07 | 2024-04-26 | 北京淇瑀信息科技有限公司 | User policy triggering method and device and electronic equipment |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108090510A (en) * | 2017-12-15 | 2018-05-29 | 北京大学 | A kind of integrated learning approach and device based on interval optimization |
| CN111080442A (en) * | 2019-12-21 | 2020-04-28 | 湖南大学 | Credit scoring model construction method, device, equipment and storage medium |
| CN111325285A (en) * | 2020-03-10 | 2020-06-23 | 五邑大学 | Fatigue driving prediction method and device based on automatic machine learning and storage medium |
-
2020
- 2020-07-21 CN CN202010703942.1A patent/CN111882426B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108090510A (en) * | 2017-12-15 | 2018-05-29 | 北京大学 | A kind of integrated learning approach and device based on interval optimization |
| CN111080442A (en) * | 2019-12-21 | 2020-04-28 | 湖南大学 | Credit scoring model construction method, device, equipment and storage medium |
| CN111325285A (en) * | 2020-03-10 | 2020-06-23 | 五邑大学 | Fatigue driving prediction method and device based on automatic machine learning and storage medium |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112836742A (en) * | 2021-02-02 | 2021-05-25 | 中国工商银行股份有限公司 | System resource adjusting method, device and equipment |
| CN112836750A (en) * | 2021-02-03 | 2021-05-25 | 中国工商银行股份有限公司 | System resource allocation method, device and equipment |
| CN112767167A (en) * | 2021-02-07 | 2021-05-07 | 中国工商银行股份有限公司 | Investment transaction risk trend prediction method and device based on ensemble learning |
| CN112767167B (en) * | 2021-02-07 | 2024-04-19 | 中国工商银行股份有限公司 | Investment transaction risk trend prediction method and device based on ensemble learning |
| CN112508698B (en) * | 2021-02-07 | 2024-04-26 | 北京淇瑀信息科技有限公司 | User policy triggering method and device and electronic equipment |
| CN113299346A (en) * | 2021-04-01 | 2021-08-24 | 腾讯科技（深圳）有限公司 | Classification model training and classifying method and device, computer equipment and storage medium |
| CN113299346B (en) * | 2021-04-01 | 2022-03-29 | 腾讯科技（深圳）有限公司 | Classification model training and classifying method and device, computer equipment and storage medium |
| CN113592160A (en) * | 2021-07-13 | 2021-11-02 | 北京淇瑀信息科技有限公司 | Fusion processing-based loss of contact risk prediction method and system for user equipment and computer equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111882426B (en) | 2024-02-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111882426B (en) | Method, device, equipment and storage medium for training business risk classifier | |
| WO2022051856A9 (en) | Method and system for training a neural network model using adversarial learning and knowledge distillation | |
| US20200175403A1 (en) | Systems and methods for expediting rule-based data processing | |
| CN107122327A (en) | The method and training system of a kind of utilization training data training pattern | |
| CN110097450A (en) | Vehicle borrows methods of risk assessment, device, equipment and storage medium | |
| CN111383030B (en) | Transaction risk detection method, device and equipment | |
| CN112015909B (en) | Knowledge graph construction method and device, electronic equipment and storage medium | |
| US11783221B2 (en) | Data exposure for transparency in artificial intelligence | |
| CN114519376A (en) | Data segmentation using neural networks | |
| CN114254762A (en) | Interpretable machine learning model construction method and device and computer equipment | |
| US11854018B2 (en) | Labeling optimization through image clustering | |
| Zaffar et al. | Comparing the performance of FCBF, Chi-Square and relief-F filter feature selection algorithms in educational data mining | |
| US20220084034A1 (en) | Encoding multi-dimensional information for deep vision learning | |
| CN113506164B (en) | Wind control decision method and device, electronic equipment and machine-readable storage medium | |
| CN115049484A (en) | Risk account prediction method and device and electronic equipment | |
| Chen et al. | Evaluation of customer behaviour with machine learning for churn prediction: The case of bank customer churn in europe | |
| Al-Mansouri et al. | Using artificial neural networks and sentiment analysis to predict upward movements in stock price | |
| CN111738864A (en) | Method, device and equipment for identifying group to which business entity belongs | |
| Selemela et al. | The role of the fourth industrial revolution in achieving economic development: challenges and opportunities | |
| CN111581068A (en) | Terminal workload calculation method and device, storage medium, terminal and cloud service system | |
| CN112884028A (en) | System resource adjusting method, device and equipment | |
| CN112150276A (en) | Training method, using method, device and equipment of machine learning model | |
| CN111079992A (en) | Data processing method, device and storage medium | |
| CN113515383B (en) | System resource data distribution method and device | |
| Hardjono et al. | Implementation of Data Mining for Churn Prediction in Music Streaming Company Using 2020 Dataset |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |