CN110689070B

CN110689070B - Training method and device of business prediction model

Info

Publication number: CN110689070B
Application number: CN201910910970.8A
Authority: CN
Inventors: 童永康; 王萌
Original assignee: 4Paradigm Beijing Technology Co Ltd
Current assignee: 4Paradigm Beijing Technology Co Ltd
Priority date: 2019-09-25
Filing date: 2019-09-25
Publication date: 2023-04-07
Anticipated expiration: 2039-09-25
Also published as: CN110689070A

Abstract

The invention discloses a method and a device for training a business prediction model, which relate to the technical field of data processing, and the main technical scheme comprises the following steps: acquiring a training sample data set of a specified service, wherein each training sample data in the training sample data set has an original label value thereof; the specified service is at least: predicting sales volume of retail goods, predicting overdue parameter values or income parameter values of bank lending products and predicting income parameter values of financial products; determining two classification label values of each training sample data based on the original label value and the label threshold value of each training sample data; performing two-classification training based on each training sample data in the training sample data set and the two-classification label values corresponding to the training sample data set to obtain two-classification models; performing regression training on the basis of all training sample data with first secondary classification label values in the training sample data set and original label values corresponding to the training sample data to obtain a regression model; and combining the two classification models and the regression model to obtain a service prediction model.

Description

Training method and device of business prediction model

Technical Field

The invention relates to the technical field of data processing, in particular to a method and a device for training a business prediction model.

Background

With the continuous development of data processing technology, data prediction in business prediction scenes such as forecasting profit parameter values of financial products can forecast a large amount of valuable or non-valuable data, so that the business prediction scenes become one of main data processing scenes.

At present, the indexes commonly used for data prediction under the business prediction scene such as the income parameter value of the financial product at least comprise: mean absolute error, root mean square error, mean absolute percent error, and symmetric mean absolute percent error. When indexes such as average absolute error, root mean square error and the like are directly used, the indexes can change along with the change of the label value of the data to be predicted of the sample, and large errors occur in data prediction. Once a large number of income parameter values of data to be predicted of the financial product have zero values, the indexes including average absolute percentage error, symmetric average absolute percentage error and the like are directly used, the problem of zero-value denominator occurs, and the used indexes lose prediction significance. Therefore, in the existing mode, the data prediction accuracy under service prediction scenes such as prediction of income parameter values of financial products is low.

Disclosure of Invention

In view of this, the invention provides a method and a device for training a business prediction model, and mainly aims to improve the accuracy of data prediction in a business prediction scene.

In a first aspect, the present invention provides a method for training a business prediction model, where the method includes:

acquiring a training sample data set of a specified service, wherein the training sample data set comprises a plurality of training sample data, and each training sample data has an original label value; wherein, the specified service is at least any one of the following: predicting sales volume of retail goods, predicting overdue parameter values or income parameter values of bank lending products and predicting income parameter values of financial products;

determining two classification label values of each training sample data based on the original label value and a preset label threshold value of each training sample data;

performing two-classification training based on each training sample data in the training sample data set and the two-classification label values corresponding to the training sample data set to obtain two-classification models; performing regression training based on the training sample data with the first secondary classification label value in the training sample data set and the original label values corresponding to the training sample data to obtain a regression model;

and combining the two classification models and the regression model to obtain a service prediction model.

In a second aspect, the present invention provides a method for predicting traffic, including:

predicting each data to be predicted in the data set to be predicted of the specified service by adopting a service prediction model aiming at the specified service to obtain a predicted value of each data to be predicted; the service prediction model is formed by combining a binary classification model and a regression model;

and predicting the appointed service based on the prediction value of each data to be predicted.

In a third aspect, the present invention provides a device for training a business prediction model, including:

the system comprises an acquisition unit, a processing unit and a processing unit, wherein the acquisition unit is used for acquiring a training sample data set of a specified service, the training sample data set comprises a plurality of training sample data, and each training sample data has an original label value; wherein the specified service is at least any one of the following: predicting sales volume of retail goods, overdue parameter values or income parameter values of bank lending products and income parameter values of financial products;

the determining unit is used for determining two classification label values of each training sample data based on the original label value and the preset label threshold value of each training sample data;

the training unit is used for performing two-classification training on the basis of each training sample data in the training sample data set and the corresponding two-classification label values thereof to obtain two-classification models; performing regression training based on the training sample data with the first secondary classification label value in the training sample data set and the original label values corresponding to the training sample data to obtain a regression model;

and the combination unit is used for combining the two classification models and the regression model to obtain a service prediction model.

In a fourth aspect, the present invention provides a service prediction apparatus, including:

the prediction unit is used for predicting each data to be predicted in the data set to be predicted of the specified service by adopting a service prediction model aiming at the specified service to obtain a prediction value of each data to be predicted; the service prediction model is formed by combining a two-classification model and a regression model;

and the prediction unit is used for predicting the specified service based on the prediction value of each data to be predicted.

In a fifth aspect, the present invention provides a computer-readable storage medium, where the storage medium includes a stored program, and when the program runs, the apparatus on which the storage medium is located is controlled to execute the method for training a traffic prediction model according to the first aspect, or execute the method for traffic prediction according to the second aspect.

In a sixth aspect, the present invention provides a storage management apparatus, including:

a memory for storing a program;

a processor, coupled to the memory, for executing the program to perform the method for training the traffic prediction model according to the first aspect, or to perform the method for traffic prediction according to the second aspect.

By means of the technical scheme, the training method and the device for the business prediction model, provided by the invention, firstly determine the two-class label values of each training sample data based on the original label value and the preset label threshold value of each training sample data of a specified business under specified business scenes of predicting the sales volume of retail commodities, predicting overdue parameter values or income parameter values of bank loan products, predicting income parameter values of financial products and the like. Then, performing two-class training based on each training sample data in the training sample data set and the corresponding two-class label value thereof to obtain two-class models, performing regression training based on each training sample data with the first two-class label value in the training sample data set and the corresponding original label value thereof to obtain regression models, and finally combining the two-class models and the regression models to obtain a service prediction model. Therefore, the scheme provided by the invention can effectively reduce the training difficulty of the business prediction model, enhance the prediction effect of the whole business prediction model and improve the accuracy of data prediction in a business prediction scene.

The foregoing description is only an overview of the technical solutions of the present invention, and the embodiments of the present invention are described below in order to make the technical means of the present invention more clearly understood and to make the above and other objects, features, and advantages of the present invention more clearly understandable.

Drawings

In order to more clearly illustrate the embodiments or technical solutions of the present invention, the drawings used in the embodiments or technical solutions in the prior art are briefly introduced below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and it is obvious for those skilled in the art that other drawings can be obtained according to these drawings without creative efforts.

FIG. 1 is a flow chart illustrating a method for training a business prediction model according to an embodiment of the present invention;

FIG. 2 is a flow chart illustrating a method for training a business prediction model according to another embodiment of the present invention;

fig. 3 is a flowchart illustrating a traffic prediction method according to an embodiment of the present invention;

FIG. 4 is a flow chart illustrating a traffic prediction method according to another embodiment of the present invention;

FIG. 5 is a schematic structural diagram of a training apparatus for a business prediction model according to an embodiment of the present invention;

FIG. 6 is a schematic structural diagram of a training apparatus for a business prediction model according to another embodiment of the present invention;

fig. 7 is a schematic structural diagram illustrating a traffic prediction apparatus according to an embodiment of the present invention;

fig. 8 is a schematic structural diagram illustrating a traffic prediction apparatus according to another embodiment of the present invention.

Detailed Description

Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be embodied in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

As shown in fig. 1, an embodiment of the present invention provides a method for training a business prediction model, where the method mainly includes:

101. acquiring a training sample data set of a specified service, wherein the training sample data set comprises a plurality of training sample data, and each training sample data has an original label value thereof; wherein, the specified service is at least any one of the following: forecasting sales of retail goods, forecasting overdue parameter values or income parameter values of bank loan products, and forecasting income parameter values of financial products.

In practical applications, the designated traffic is at least traffic in which the label value of the sample data in the traffic prediction scenario contains a large number of zero values and values close to zero. Optionally, the designated service is at least any one of the following: forecasting sales of retail goods, forecasting overdue parameter values or income parameter values of bank loan products, and forecasting income parameter values of financial products.

In this embodiment, the training sample data set may be stored in a preset storage location, and when the training sample data set needs to be acquired, the training sample data set may be acquired from the storage location through a preset interface. Optionally, the predetermined interface may include, but is not limited to, an API interface. It should be noted that, in order to ensure that the training sample data set can reflect the real data of the specified service to the maximum extent, each training sample data in the training sample data set in the storage location is updated at a preset frequency, so that the latest service prediction model meeting the specified service can be trained in time based on the updated training sample data set. In addition, the number of training sample data in the training sample data set is not specifically limited in this embodiment, and the number of training sample data may be determined based on the training speed and/or the training precision required by the service.

In this embodiment, the training sample data is structured data, and the training sample data in different specified services has different structures. Illustratively, when the designated business is a profit parameter value of the predicted financial product, at least the sold time length, the sold share, the historical profit, the profit amount and the profit rate are included in the training sample data.

In this embodiment, each training sample data has its own original tag value, and the original tag value is a parameter value that needs to be predicted in a specific service, that is, the original tag value is a variable that is desired to be predicted. For example, when the designated service is the predicted sales volume of the retail goods, the original label value of each training sample data is the sales volume of the retail goods. When the appointed business is used for predicting overdue parameter values of bank loan products, the original label value of each training sample data is the overdue parameter value of the bank loan product, wherein the overdue parameter value of the bank loan product is at least any one of the following: amount of overdue and overdue rate. When the appointed business is the income parameter value of the forecast bank loan product, the original label value of each training sample data is the income parameter value of the bank loan product, wherein the income parameter value of the bank loan product is at least any one of the following: the amount of revenue and the rate of revenue. When the appointed business is a profit parameter value of a predicted financial product, the original label value of each training sample data is the profit parameter value of the financial product, wherein the profit parameter value of the financial product is at least any one of the following: the amount of revenue and the rate of revenue.

102. And determining the two classification label values of each training sample data based on the original label value and the preset label threshold value of each training sample data.

In this embodiment, the two classification label values of each training sample data are determined according to the size relationship between the original label value of each training sample data and the preset label threshold value. Specifically, the two-class label value of each training sample data with the original label value equal to or greater than the preset label threshold value is determined as the first two-class label value. And determining the two-classification label values of the training sample data with the original label value smaller than the preset label threshold value as second two-classification label values. The first and second classification labels are preset first threshold values, and the second classification label is a preset second threshold value. It should be noted that the first threshold and the second threshold are different, and specific values of the first threshold and the second threshold may be flexibly set based on the service requirement.

For example, if the business is specified as predicting the profit amount of the financial product, and the preset tag threshold is 100, the two-class tag value of the training sample data with the profit amount greater than or equal to the tag threshold "100" is determined as the first two-class tag value "1". The binary label value of the training sample data with the profit amount smaller than the label threshold value of "100" is determined as the second binary label value of "0".

103. Performing two-classification training based on each training sample data in the training sample data set and the two-classification label values corresponding to the training sample data set to obtain two-classification models; and performing regression training based on the training sample data with the first secondary classification label value in the training sample data set and the original label value corresponding to the training sample data to obtain a regression model.

In practical applications, the process of training the two-class model and the process of training the regression model may be performed sequentially, for example, the process of training the two-class model is performed first and then the process of training the regression model is performed, or the process of training the regression model is performed first and then the process of training the two-class model is performed. Of course, in order to shorten the training time of the traffic prediction model, the process of training the binary model and the process of training the regression model may be performed synchronously.

The following describes a process of performing two-class training to obtain two-class models based on each training sample data in the training sample data set and its corresponding two-class label values: selecting a two-classification training method based on two-classification training requirements and/or data characteristics of each training sample data, wherein the classification training requirements comprise training efficiency and/or training precision; the data characteristics include at least one of: the total amount of data, the number of continuous variables, and the number of discrete variables for each training sample data. And performing two-classification training on each training sample data in the training sample data set and the corresponding two-classification label values thereof by using the selected two-classification training method to obtain two-classification models.

Specifically, when the number of continuous variables in each training sample data is greater than a preset continuous variable threshold, the selected binary training method is a gradient boosting decision tree GBDT method. And when the number of the discrete variables in each training sample data is greater than a preset discrete variable threshold value, selecting the binary training method as a logistic regression method. And when the total data amount of each training sample data is larger than a preset data amount threshold value, selecting a two-classification training method as a neural network method.

Specifically, the following describes how to perform regression training on each training sample data having the first secondary class label value in the training sample data set and the original label value corresponding to each training sample data to obtain a regression model: selecting a regression training method based on regression training requirements and/or data characteristics of each training sample data with a first secondary classification label value, wherein the regression training requirements comprise training efficiency and/or training precision; the data characteristics include at least one of: the total amount of data, the number of continuous variables, and the number of discrete variables for each training sample data. And performing regression training on each training sample data with the first secondary classification label value in the training sample data set and the original label value corresponding to each training sample data to obtain a regression model by using the selected regression training method.

Specifically, when the number of continuous variables in each training sample data with the first secondary classification label value is greater than a preset continuous variable threshold, the selected regression training method is a gradient regression tree GBRT method. And when the total data amount of each training sample data with the first secondary classification label value is larger than a preset data amount threshold value, the selected regression training method is a neural network method. When the training efficiency is greater than a preset efficiency threshold value, the selected regression training method is a linear regression method.

104. And combining the classification model and the regression model to obtain a service prediction model.

In this embodiment, the process of combining the classification model and the regression model to obtain the service prediction model at least includes: and testing the binary classification model and the regression model by using each test sample data in the test sample data set of the specified service. And combining the two classification models and the regression model which pass the test to obtain a service prediction model.

Specifically, in order to ensure the prediction effect of the service prediction model, the classification model and the regression model cannot be directly combined, but the two classification models and the regression model need to be tested by using each test sample data in the test sample data set of the specified service, and after the tests of the two classification models and the regression model pass, the two classification models and the regression model can be combined to obtain the service prediction model. It should be noted that, when the tests of the two classification models and the regression model are failed, it is indicated that the prediction effects of the two classification models and the regression model are poor, new training sample data needs to be obtained again or the hyper-parameters of the training classification models and/or the regression model need to be adjusted again, new classification models and regression models are obtained through retraining, the above-mentioned test process is repeated, and the two classification models and the regression model which pass the tests are not combined to obtain the service prediction model until the tests of the two classification models and the regression model which pass the tests are passed.

Specifically, the specific process of obtaining the service prediction model by combining the two classification models and the regression model that pass the test may be at least: determining the execution logics of the two-classification model and the regression model, and combining the two-classification model and the regression model based on the execution logics, wherein the execution logics are that the two-classification model predicts sample data to be predicted firstly, and the regression model predicts the data with a specific label value based on the prediction of the two-classification model, namely the regression model selectively predicts the sample data to be predicted.

According to the training method of the business prediction model provided by the embodiment of the invention, firstly, under the specified business scenes of predicting sales volume of retail commodities, predicting overdue parameter values or income parameter values of bank loan products, predicting income parameter values of financial products and the like, and based on the original label value and the preset label threshold value of each training sample data of specified business, the binary label value of each training sample data is determined. Then, performing two-class training based on each training sample data in the training sample data set and the corresponding two-class label value thereof to obtain two-class models, performing regression training based on each training sample data with the first two-class label value in the training sample data set and the corresponding original label value thereof to obtain regression models, and finally combining the two-class models and the regression models to obtain a service prediction model. Therefore, the scheme provided by the embodiment of the invention can effectively reduce the training difficulty of the business prediction model, enhance the prediction effect of the whole business prediction model and improve the accuracy of data prediction in a business prediction scene.

Further, according to the method shown in fig. 1, another embodiment of the present invention further provides a method for training a business prediction model, as shown in fig. 2, the method mainly includes:

201. acquiring a training sample data set of a specified service, wherein the training sample data set comprises a plurality of training sample data, and each training sample data has an original label value thereof; wherein the specified service is at least any one of the following: forecasting sales of retail goods, forecasting overdue parameter values or income parameter values of bank loan products, and forecasting income parameter values of financial products.

The detailed description of this step is based on the same basic steps as step 101 described above, and therefore will not be described here.

202. And determining the two-class label value of the training sample data of which the original label value is equal to or greater than a preset label threshold value in the training sample data set as a first two-class label value, wherein the first two-class label value is a preset first threshold value.

For example, if the business is specified as predicting the profit amount of the financial product, and the preset tag threshold is 100, the two-class tag value of the training sample data with the profit amount greater than or equal to the tag threshold "100" is determined as the first two-class tag value "1".

203. Determining the two classification label values of the training sample data of which the original label values are smaller than a preset label threshold value in the training sample data set as second two classification label values, wherein the second two classification label values are preset second threshold values, and the second threshold values are different from the first threshold values.

For example, if the business is specified as predicting the profit amount of the financial product, and the preset tag threshold is 100, the two-class tag value of the training sample data with the profit amount smaller than the tag threshold "100" is determined as the second two-class tag value "0".

204. Performing two-classification training based on each training sample data in the training sample data set and the two-classification label values corresponding to the training sample data set to obtain two-classification models; and performing regression training based on the training sample data with the first secondary classification label value in the training sample data set and the original label values corresponding to the training sample data to obtain a regression model.

The detailed description of this step is based on the same basic steps as step 103 described above, and therefore will not be described here.

205. And determining the predicted value of each test sample data in the test sample data set of the specified service by using the two classification models and the regression model.

In practical application, each test sample data has a determined original label value, and the predicted value of the test sample data is determined by using the two-classification model and the regression model, namely, the original label value is predicted so as to predict the prediction effect of the two-classification model and the regression model based on the predicted value and the original label value.

The following describes a specific process for determining the predicted value of each test sample data in the test sample data set of the specified service by using the binary classification model and the regression model: predicting each test sample data by adopting a two-classification model to obtain a two-classification label value of each test sample data; determining the predicted value of each test sample data with the second classification label value as the corresponding second classification label value; and predicting the test sample data with the two-classification label value as the first second-class label value by adopting the regression model to obtain the predicted value of the test sample data with the two-classification label value as the first second-class label value. It should be noted that, in the whole process, the non-zero value sample data is judged by predicting through the two classification models, and then the regression prediction is performed on the non-zero value sample data predicted by the two classification models through the regression model, so that the training difficulty of the regression model is effectively reduced, and the overall model effect is enhanced.

Illustratively, a binary model is adopted to predict each test sample data to obtain a binary label value of each test sample data, and the predicted value of each test sample data with the binary label value being a second binary label value of "0" is determined to be 0. And then, predicting each test sample data with the two-classification label value of 1 by adopting a regression model to obtain the predicted value of each test sample data with the two-classification label value of 1.

206. And determining a model prediction index according to the predicted value of each test sample data.

Specifically, the process of determining the model prediction index according to the predicted value of each test sample data at least includes the following two processes:

firstly, according to the predicted value of each test sample data, determining the root mean square error aiming at each test sample data; and determining the root mean square error of each test sample data as a model prediction index.

Specifically, the root mean square error for each test sample data is calculated by formula (2).

Wherein A characterizes the model predictor; the D characterizes the root mean square error; the n represents the total amount of each test sample data; y is _j Representing the predicted value of the jth test sample data; the above-mentioned

And characterizing the original label value of the jth test sample data, namely the true value.

Secondly, determining the average absolute error aiming at each test sample data according to the predicted value of each test sample data; and determining the average absolute error of each test sample data as the model prediction index.

Specifically, the average absolute error for each test sample data is calculated by formula (3).

Wherein A characterizes the model predictor; said E characterizes said mean absolute error; the n represents the total amount of each test sample data; y is _j Characterizing a predicted value of jth test sample data; the above-mentioned

And representing the original label value of the jth test sample data, namely the true value.

It should be noted that, by using the abstract absolute value indexes such as the model evaluation indexes such as the mean absolute error or the root mean square error, the intuitive percentage indexes such as the mean absolute percentage error MAPE and the symmetric mean absolute percentage error SMAPE can be completely avoided, and when more zero values exist in a sample, the problem of zero-value denominator is encountered.

207. And determining at least one comparative prediction index based on the original label value of each test sample data in the test data set.

In this embodiment, the number of comparison prediction indexes may be flexibly determined based on the service requirement. It should be noted that, in order to reduce the probability of the test deviation between the two-classification model and the regression model, a larger number of comparison prediction indexes may be selected.

In this embodiment, the process of determining the comparative prediction index based on the original label value of each test sample data in the test data set at least includes the following two processes:

first, the mean of the original label values of each test data is determined as a comparative predictor.

Second, a mode in the original label value of each test data is determined as a comparative prediction index.

208. Determining whether the binary model and the regression model pass the test based on the at least one comparison predictor and the model predictor.

In this embodiment, the process of determining whether the binary model and the regression model pass the test based on the at least one comparison predictor and the model predictor at least includes the following two processes:

first, the difference between the model prediction index and each comparison prediction index is calculated, wherein each prediction index has a corresponding preset difference range. And if the difference value between the model prediction index and each comparison prediction index is within the preset difference value range corresponding to each comparison prediction index, determining that the two classification models and the regression model pass the test.

Specifically, when the difference between the model prediction index and each comparative prediction index is within the preset difference range corresponding to each comparative prediction index, it is indicated that the model prediction index is very close to each comparative prediction index, and the prediction effects of the two classification models and the regression model are good, and it is determined that the two classification models and the regression model pass the test.

Specifically, if at least one difference value between the model prediction index and each comparison prediction index is not within the corresponding preset difference value range, which indicates that the prediction effect of the two-classification model and the regression model is poor, it is determined that the tests of the two-classification model and the regression model do not pass, and the two-classification model and the regression model need to be retrained.

Secondly, calculating an error index of the model prediction index and each comparison prediction index through formula (1) based on the at least one comparison prediction index and the model prediction index; and if the error indexes of the model prediction indexes and the comparative prediction indexes are both larger than the preset error index threshold corresponding to the comparative prediction indexes, determining that the binary models and the regression models pass the test.

Wherein, T is _i Representing an error index of the model prediction index and the ith comparison prediction index; the A represents the model predictor; b is described _i Characterizing the ith comparison predictor;

specifically, abstract absolute value indexes such as model evaluation indexes such as average absolute errors or root mean square errors can be converted into visual percentage indexes through the formula (1). When the model prediction index A is larger than the comparison prediction index B _i Time, error index T _i The value is negative, which indicates that the prediction capability of the combined model formed by combining the two classification models and the regression model is weaker than that of the comparison index. When the model prediction index A is smaller than the comparison prediction index B _i Time, error index T _i The value is positive, at this time, the prediction capability of the combined model formed by combining the two classification models and the regression model is stronger than that of the contrast index, and the error index T _i The closer to 100%, the better the prediction effect of the characterization combination model. Error index T _i The 0 percent represents that the combination model formed by combining the binary model and the regression model is not improved compared with a contrast index, and an error index T _i The future is perfectly predicted for a combined model formed by combining a 100% representative binary model and a regression model. Therefore, the larger the error index is, the better the prediction effect of the binary model and the regression model is.

209. And combining the two classification models and the regression model which pass the test to obtain the service prediction model.

The detailed description of this step is based on the same as that of step 104 described above, and therefore will not be described here.

As shown in fig. 3, an embodiment of the present invention provides a service prediction method, where the method mainly includes:

301. predicting each data to be predicted in the data set to be predicted of the specified service by adopting a service prediction model aiming at the specified service to obtain a predicted value of each data to be predicted; the service prediction model is formed by combining a two-classification model and a regression model.

In practical application, the business prediction model of the designated business is formed by combining a classification model and a regression model, and when the business prediction model is adopted to predict data to be predicted, the data to be predicted is predicted through a two-classification model so as to judge the data to be predicted with a two-classification label value as a non-zero label value. And then, carrying out regression prediction on the data to be predicted with the non-zero tag value through a regression model, thereby effectively reducing the training difficulty of the regression model and enhancing the effect of the whole service prediction model.

Further, in order to ensure the prediction effect of the service prediction model, the latest service data can be used as training sample data to train to obtain the latest service prediction model.

302. And predicting the appointed service based on the prediction value of each data to be predicted.

In this embodiment, the assigned service is predicted based on the prediction value of each piece of data to be predicted, and actually, the distribution of the prediction values is given, so that the user adjusts the assigned service based on the distribution.

Illustratively, in a financial product income forecasting scene of a large bank, the forecasting value of each data to be forecasted is a forecasting value of income amount for 6 months continuously. And when the predicted value of the income amount is smaller than the preset expected value, giving a prompt of poor income of the financial product.

The service prediction method provided by the embodiment of the invention is obtained by the service prediction model which consists of a binary classification model for performing binary prediction on data and a regression model for performing regression prediction on the data with the characteristic classification label values, so that the service prediction model can perform prediction classification on each data to be predicted through the binary classification model firstly, and then performs regression prediction on the data to be predicted with the specific label values through the regression model.

Further, according to the method shown in fig. 3, another embodiment of the present invention further provides a traffic prediction method, as shown in fig. 4, the method mainly includes:

401. and predicting the data to be predicted by adopting the two classification models to obtain two classification label values of the data to be predicted.

402. And determining the predicted value of each data to be predicted, which takes the binary label value as the second binary label value, as the second binary label value corresponding to each predicted value.

403. And predicting the data to be predicted with the two-classification label value as the first second-class label value by adopting the regression model to obtain the predicted value of the data to be predicted with the two-classification label value as the first second-class label value.

Illustratively, the process examples of the above steps 401 to 403 are: and predicting the data to be predicted by adopting a binary model to obtain a binary label value of the data to be predicted, and determining the predicted value of the data to be predicted, which has the binary label value of 0, as 0. And then, predicting the data to be predicted with the two-classification label value as the first two-classification label value of 1 by adopting a regression model to obtain the predicted value of the data to be predicted with the two-classification label value of 1.

404. And predicting the appointed service based on the prediction value of each data to be predicted.

Further, according to the embodiment of the training method of the business prediction model, another embodiment of the present invention further provides a training apparatus of a business prediction model, as shown in fig. 5, the apparatus includes:

an obtaining unit 51, configured to obtain a training sample data set of a specified service, where the training sample data set includes multiple training sample data, and each training sample data has its own original tag value; wherein the specified service is at least any one of the following: predicting sales volume of retail goods, predicting overdue parameter values or income parameter values of bank lending products and predicting income parameter values of financial products;

a determining unit 52, configured to determine two classification tag values of each training sample data based on the original tag value and a preset tag threshold of each training sample data;

a training unit 53, configured to perform two-class training based on each training sample data in the training sample data set and its respective corresponding two-class label value, to obtain two-class models; performing regression training based on the training sample data with the first secondary classification label value in the training sample data set and the original label values corresponding to the training sample data to obtain a regression model;

and the combining unit 54 is configured to combine the classification model and the regression model to obtain a service prediction model.

According to the training device of the business prediction model provided by the embodiment of the invention, firstly, under the specified business scenes of predicting sales volume of retail commodities, predicting overdue parameter values or income parameter values of bank loan products, predicting income parameter values of financial products and the like, the binary label values of training sample data are determined based on the original label values and the preset label threshold values of the training sample data of the specified business. Then, performing two-class training based on each training sample data in the training sample data set and the corresponding two-class label value thereof to obtain two-class models, performing regression training based on each training sample data with the first two-class label value in the training sample data set and the corresponding original label value thereof to obtain regression models, and finally combining the two-class models and the regression models to obtain a service prediction model. Therefore, the scheme provided by the embodiment of the invention can effectively reduce the training difficulty of the business prediction model, enhance the prediction effect of the whole business prediction model and improve the accuracy of data prediction in a business prediction scene.

Optionally, the original label value of each training sample data is a parameter value to be predicted in the specified service;

when the designated business is the sales volume of the predicted retail commodity, the original label value of each training sample data is the sales volume of the retail commodity;

when the designated business is used for predicting overdue parameter values of bank loan products, the original label value of each training sample data is the overdue parameter value of the bank loan product, wherein the overdue parameter value of the bank loan product is at least any one of the following values: amount and rate of overdue;

when the designated business is a revenue parameter value of a predicted bank loan product, the original label value of each training sample data is the revenue parameter value of the bank loan product, wherein the revenue parameter value of the bank loan product is at least any one of the following values: income amount and income rate;

when the designated business is a profit parameter value of a predicted financial product, the original tag value of each training sample data is the profit parameter value of the financial product, wherein the profit parameter value of the financial product is at least any one of the following: the amount of revenue and the rate of revenue.

Optionally, as shown in fig. 6, the combining unit 54 includes:

a test module 541, configured to test the binary classification model and the regression model by using each test sample data in the test sample data set of the specified service;

and the combination module 542 is used for combining the two classification models and the regression model which pass the test to obtain the service prediction model.

Optionally, as shown in fig. 6, the test module 541 includes:

a first determining submodule 5411, configured to determine a predicted value of each piece of test sample data by using the two classification models and the regression model; determining a model prediction index according to the predicted value of each test sample data;

a second determining submodule 5412, configured to determine at least one comparison prediction index based on an original tag value of each test sample data in the test data set; determining whether the binary model and the regression model pass the test based on the at least one comparison predictor and the model predictor.

Optionally, the first determining submodule 5411 is configured to predict each test sample data by using the two classification models, so as to obtain two classification label values of each test sample data; determining the predicted value of each test sample data with the second classification label value as the corresponding second classification label value; and predicting each test sample data with the first second-class label value as the two-class label value by adopting the regression model to obtain the predicted value of each test sample data with the first second-class label value as the two-class label value.

Optionally, the first determining sub-module 5411 is configured to determine, according to the predicted value of each test sample data, an average absolute error or a root-mean-square error for each test sample data; and determining the average absolute error or the root mean square error of each test sample data as the model prediction index.

Optionally, the second determining sub-module is configured to determine a mean of the original tag values of each test data as a comparison prediction index, and/or determine a mode of the original tag values of each test data as the comparison prediction index.

Optionally, as shown in fig. 6, the second determining submodule 5412 is configured to determine that the two classification models and the regression model pass the test if the difference between the model prediction index and each comparison prediction index is within a preset difference range corresponding to each comparison prediction index.

Optionally, as shown in fig. 6, the second determining sub-module 5412 is configured to calculate, based on the at least one comparison predictor and the model predictor, an error indicator of the model predictor and each comparison predictor by formula (1); if the error indexes of the model prediction indexes and the comparative prediction indexes are both larger than the preset error index threshold corresponding to the comparative prediction indexes, determining that the binary model and the regression model pass the test;

wherein, the T is _i Representing an error index of the model prediction index and the ith comparison prediction index; the A represents the model predictor; b is _i And characterizing the ith comparison predictor.

Optionally, as shown in fig. 6, the determining unit 52 includes:

a first determining module 521, configured to determine a first secondary class label value of a two-class label value of training sample data in the training sample data set, where an original label value of the training sample data set is equal to or greater than a preset label threshold, where the first secondary class label value is a preset first threshold;

a second determining module 522, configured to determine a second classification tag value of the training sample data in which a label value of the original tag value in the training sample data set is smaller than a preset tag threshold, where the second classification tag value is a preset second threshold, and the second threshold is different from the first threshold.

Optionally, as shown in fig. 6, the training unit 53 includes:

the first training module 531 is configured to select a two-classification training method based on two-classification training requirements and/or data characteristics of each training sample data, where the two-classification training requirements include training efficiency and/or training accuracy; the data characteristics include at least one of: the total data amount, the number of continuous variables and the number of discrete variables of each training sample data; and performing two-classification training on each training sample data in the training sample data set and the two-classification label values corresponding to the training sample data set by using the selected two-classification training method.

Optionally, when the number of continuous variables in each training sample data is greater than a preset continuous variable threshold, the selected binary training method is a gradient boosting decision tree GBDT method;

when the number of discrete variables in each training sample data is larger than a preset discrete variable threshold value, the selected binary training method is a logistic regression method;

and when the total data amount of each training sample data is larger than a preset data amount threshold value, selecting a two-classification training method as a neural network method.

Optionally, as shown in fig. 6, the training unit 53 includes:

a second training module 532, configured to select a regression training method based on a regression training requirement and/or a data feature of each training sample data with a first secondary class label value, where the regression training requirement includes training efficiency and/or training accuracy; the data characteristics include at least one of: the total data amount, the number of continuous variables and the number of discrete variables of each training sample data; and performing regression training on each training sample data with the first secondary classification label value in the training sample data set and the original label value corresponding to each training sample data by using the selected regression training method.

Optionally, when the number of continuous variables in each training sample data with the first secondary classification label value is greater than a preset continuous variable threshold, the selected regression training method is a gradient regression tree GBRT method;

when the total data amount of each training sample data with the first secondary classification label value is larger than a preset data amount threshold value, the selected regression training method is a neural network method;

when the training efficiency is greater than a preset efficiency threshold value, the selected regression training method is a linear regression method.

In the training device for the service prediction model provided in the embodiment of the present invention, for details of methods used in the operation process of each function module, reference may be made to the corresponding methods in the embodiments of the methods in fig. 1 and fig. 2, and details are not described here again.

Further, according to the embodiment of the foregoing service prediction method, another embodiment of the present invention further provides a service prediction apparatus, as shown in fig. 7, the apparatus includes:

the prediction unit 61 is configured to predict, by using a service prediction model for a specific service, each piece of data to be predicted in a set of data to be predicted of the specific service, so as to obtain a predicted value of each piece of data to be predicted; the business prediction model is formed by combining a classification model and a regression model;

and the predicting unit 62 is configured to predict the specified service based on a prediction value of each piece of data to be predicted.

The service prediction device provided by the embodiment of the invention is obtained by the service prediction model which consists of a binary classification model for performing binary prediction on data and a regression model for performing regression prediction on the data with the characteristic classification label values, so that the service prediction model can perform prediction classification on each data to be predicted through the binary classification model firstly, and then performs regression prediction on the data to be predicted with a specific label value through the regression model.

Optionally, as shown in fig. 8, the prediction unit 61 includes:

the first prediction module 611 is configured to predict each data to be predicted by using the two classification models, so as to obtain two classification tag values of each data to be predicted;

a second prediction module 612, configured to determine the predicted value of each to-be-predicted data whose two-class tag value is a second two-class tag value as a second two-class tag value corresponding to the predicted value;

the third prediction module 613 is configured to predict each data to be predicted, whose two-class label value is the first two-class label value, by using the regression model, so as to obtain a predicted value of each data to be predicted, whose two-class label value is the first two-class label value.

In the service prediction apparatus provided in the embodiment of the present invention, for details of methods used in the operation process of each function module, reference may be made to the corresponding methods in the embodiments of the methods in fig. 3 and fig. 4, which are not described herein again.

Further, according to the foregoing embodiment, another embodiment of the present invention further provides a computer-readable storage medium, where the storage medium includes a stored program, and when the program runs, the apparatus on which the storage medium is located is controlled to execute the method for training the traffic prediction model described in any one of the foregoing embodiments, or execute the method for predicting traffic.

Further, according to the above embodiment, another embodiment of the present invention also provides a storage management apparatus, including:

a memory for storing a program;

a processor, coupled to the memory, for executing the program to perform the method for training the traffic prediction model according to any one of the above items, or to perform the traffic prediction method according to any one of the above items.

In the foregoing embodiments, the descriptions of the respective embodiments have respective emphasis, and for parts that are not described in detail in a certain embodiment, reference may be made to the related descriptions of other embodiments.

The embodiment of the invention discloses:

A1. a training method of a business prediction model comprises the following steps:

acquiring a training sample data set of a specified service, wherein the training sample data set comprises a plurality of training sample data, and each training sample data has an original label value; wherein the specified service is at least any one of the following: predicting sales volume of retail goods, predicting overdue parameter values or income parameter values of bank lending products and predicting income parameter values of financial products;

A2. According to the method of A1, the original label value of each training sample data is a parameter value needing to be predicted in the specified service;

A3. The method according to A1, wherein the combining the two classification models and the regression model to obtain a business prediction model includes:

testing the two classification models and the regression model by using each test sample data in the test sample data set of the specified service;

and combining the two classification models and the regression model which pass the test to obtain the service prediction model.

A4. According to the method described in A3, the testing the two classification models and the regression model by using each test sample data in the test sample data set of the specified service includes:

determining the predicted value of each test sample data by using the two classification models and the regression model;

determining a model prediction index according to the predicted value of each test sample data;

determining at least one comparison prediction index based on the original label value of each test sample data in the test data set;

determining whether the two classification models and the regression model pass the test based on the at least one comparison predictor and the model predictor.

A5. The method of A4, wherein the determining the predicted value for each test sample data using the two classification models and the regression model, comprises:

predicting each test sample data by adopting the two classification models to obtain two classification label values of each test sample data;

determining the predicted value of each test sample data with the second classification label value as the corresponding second classification label value;

and predicting the test sample data with the two-classification label value as the first second-class label value by adopting the regression model to obtain the predicted value of the test sample data with the two-classification label value as the first second-class label value.

A6. According to the method described in A4, determining a model prediction index according to the predicted value of each test sample data includes:

determining the average absolute error or the root mean square error aiming at each test sample data according to the predicted value of each test sample data;

and determining the average absolute error or the root-mean-square error of each test sample data as the model prediction index.

A7. According to the method described in A4, the determining at least one comparative prediction index based on the original label value of each test sample data in the test data set includes:

and determining the average value of the original label values of the test data as a comparison prediction index, and/or determining the mode in the original label values of the test data as the comparison prediction index.

A8. The method of A4, the determining whether the two classification models and the regression model test pass based on the at least one comparison predictor and the model predictor, comprising:

and if the difference value between the model prediction index and each comparison prediction index is within the preset difference value range corresponding to each comparison prediction index, determining that the binary model and the regression model pass the test.

A9. The method of A4, the determining whether the two classification models and the regression model test pass based on the at least one comparison predictor and the model predictor, comprising:

calculating an error indicator for the model predictor and each comparison predictor by a first formula based on the at least one comparison predictor and the model predictor;

the first formula is:

wherein, T is _i Representing an error index of the model prediction index and the ith comparison prediction index; the above-mentionedA represents the model prediction index; b is _i Characterizing the ith comparison predictor;

and if the error indexes of the model prediction indexes and the comparative prediction indexes are both larger than the preset error index threshold corresponding to the comparative prediction indexes, determining that the binary models and the regression models pass the test.

A10. According to the method described in any one of A1 to A9, the determining the two-class label value of each training sample data based on the original label value and the preset label threshold value of each training sample data includes:

determining a first second-class label value of a training sample data with an original label value equal to or greater than a preset label threshold value in the training sample data set, wherein the first second-class label value is a preset first threshold value;

determining the binary label value of the training sample data of which the original label value is smaller than a preset label threshold value in the training sample data set as a second binary label value, wherein the second binary label value is a preset second threshold value, and the second threshold value is different from the first threshold value.

A11. According to the method described in any one of A1 to A9, the performing two-class training based on each training sample data in the training sample data set and its respective corresponding two-class label value includes:

selecting a two-classification training method based on two-classification training requirements and/or data characteristics of each training sample data, wherein the classification training requirements comprise training efficiency and/or training precision; the data characteristics include at least one of: the total data amount, the number of continuous variables and the number of discrete variables of each training sample data;

and performing two-classification training on each training sample data in the training sample data set and the two-classification label values corresponding to the training sample data set by using the selected two-classification training method.

A12. According to the method A11, when the number of continuous variables in each training sample data is greater than a preset continuous variable threshold, the selected binary training method is a gradient boosting decision tree GBDT method;

A13. According to the method described in any one of A1 to A9, the performing regression training based on each training sample data having the first second-class label value in the training sample data set and the original label value corresponding to each training sample data includes:

selecting a regression training method based on regression training requirements and/or data characteristics of each training sample data with a first secondary classification label value, wherein the regression training requirements comprise training efficiency and/or training precision; the data characteristics include at least one of: the total data amount, the number of continuous variables and the number of discrete variables of each training sample data;

and performing regression training on each training sample data with the first secondary classification label value in the training sample data set and the original label value corresponding to each training sample data by using the selected regression training method.

A14. According to the method in a13, when the number of continuous variables in each piece of training sample data with the first secondary classification label value is greater than a preset continuous variable threshold, the selected regression training method is a gradient regression tree GBRT method;

B1. A traffic prediction method, comprising:

predicting each data to be predicted in the data set to be predicted of the specified service by adopting a service prediction model aiming at the specified service to obtain a predicted value of each data to be predicted; the service prediction model is formed by combining a two-classification model and a regression model;

B2. According to the method of B1, predicting, by using a service prediction model for a specific service, each data to be predicted in a data set to be predicted of the specific service to obtain a predicted value of each data to be predicted, including:

predicting the data to be predicted by adopting the two classification models to obtain two classification label values of the data to be predicted;

determining the predicted value of each data to be predicted, of which the two-classification label value is a second two-classification label value, as a second two-classification label value corresponding to the predicted value;

and predicting the data to be predicted with the two-classification label value as the first second-class label value by adopting the regression model to obtain the predicted value of the data to be predicted with the two-classification label value as the first second-class label value.

C1. An apparatus for training a traffic prediction model, comprising:

the system comprises an acquisition unit, a processing unit and a processing unit, wherein the acquisition unit is used for acquiring a training sample data set of a specified service, the training sample data set comprises a plurality of training sample data, and each training sample data has an original label value thereof; wherein the specified service is at least any one of the following: predicting sales volume of retail goods, predicting overdue parameter values or income parameter values of bank lending products and predicting income parameter values of financial products;

the training unit is used for performing two-classification training on the basis of each training sample data in the training sample data set and the two-classification label values corresponding to the training sample data set to obtain two-classification models; performing regression training based on the training sample data with the first secondary classification label value in the training sample data set and the original label values corresponding to the training sample data to obtain a regression model;

C2. According to the device of C1, the original label value of each training sample data is a parameter value needing to be predicted in the specified service;

C3. The apparatus of C1, the combining unit comprising:

the test module is used for testing the binary classification model and the regression model by using each test sample data in the test sample data set of the specified service;

and the combination module is used for combining the two classification models and the regression model which pass the test to obtain the service prediction model.

C4. The apparatus of C3, the test module comprising:

the first determining submodule is used for determining the predicted value of each test sample data by utilizing the two-classification model and the regression model; determining a model prediction index according to the predicted value of each test sample data;

the second determining submodule is used for determining at least one comparison prediction index based on the original label value of each test sample data in the test data set; determining whether the binary model and the regression model pass the test based on the at least one comparison predictor and the model predictor.

C5. According to the device of C4, the first determining submodule is configured to predict each test sample data by using the two classification models, so as to obtain a two-classification label value of each test sample data; determining the predicted value of each test sample data with the second classification label value as the corresponding second classification label value; and predicting each test sample data with the first second-class label value as the two-class label value by adopting the regression model to obtain the predicted value of each test sample data with the first second-class label value as the two-class label value.

C6. According to the apparatus described in C4, the first determining sub-module is configured to determine an average absolute error or a root mean square error for each test sample data according to the predicted value of each test sample data; and determining the average absolute error or the root-mean-square error of each test sample data as the model prediction index.

C7. The apparatus according to C4, the second determining sub-module is configured to determine a mean of the original label values of each test data as a comparison predictor, and/or determine a mode of the original label values of each test data as a comparison predictor.

C8. The device according to C4, wherein the second determining submodule is configured to determine that the two-classification model and the regression model pass the test if the difference between the model prediction index and each of the comparative prediction indexes is within a preset difference range corresponding to each of the comparative prediction indexes.

C9. The apparatus according to C4, the second determining submodule being configured to calculate, based on the at least one comparison predictor and the model predictor, an error indicator for the model predictor and each comparison predictor by a first formula; if the error indexes of the model prediction indexes and the comparative prediction indexes are both larger than the preset error index threshold corresponding to the comparative prediction indexes, determining that the binary model and the regression model pass the test;

the first formula is:

wherein, T is _i Representing an error index of the model prediction index and the ith comparison prediction index; the A represents the model predictor; b is _i And characterizing the ith comparison predictor.

C10. The apparatus according to any one of C1-C9, the determining unit comprising:

a first determining module, configured to determine a first second-class label value of a two-class label value of training sample data in the training sample data set, where an original label value of the training sample data set is equal to or greater than a preset label threshold value, where the first second-class label value is a preset first threshold value;

a second determining module, configured to determine a second classification tag value of the training sample data with the original tag value in the training sample data set smaller than a preset tag threshold, where the second classification tag value is a preset second threshold, and the second threshold is different from the first threshold.

C11. The apparatus of any of C1-C9, the training unit comprising:

the training system comprises a first training module, a second training module and a third training module, wherein the first training module is used for selecting a two-classification training method based on two-classification training requirements and/or data characteristics of each training sample data, and the two-classification training requirements comprise training efficiency and/or training precision; the data characteristics include at least one of: the total data amount, the number of continuous variables and the number of discrete variables of each training sample data; and performing two-classification training on each training sample data in the training sample data set and the two-classification label values corresponding to the training sample data set by using the selected two-classification training method.

C12. The apparatus according to the claim in the point C11,

when the number of continuous variables in each training sample data is larger than a preset continuous variable threshold, the selected binary training method is a gradient lifting decision tree (GBDT) method;

C13. The apparatus of any of C1-C9, the training unit comprising:

the second training module is used for selecting a regression training method based on regression training requirements and/or data characteristics of each training sample data with a first secondary classification label value, wherein the regression training requirements comprise training efficiency and/or training precision; the data characteristics include at least one of: the total data amount, the number of continuous variables and the number of discrete variables of each training sample data; and performing regression training on each training sample data with the first secondary classification label value in the training sample data set and the original label value corresponding to each training sample data by using the selected regression training method.

C14. According to the apparatus as set forth in C13,

when the number of continuous variables in each training sample data with the first secondary classification label value is larger than a preset continuous variable threshold value, the selected regression training method is a progressive gradient regression tree GBRT method;

D1. A traffic prediction apparatus comprising:

the prediction unit is used for predicting each data to be predicted in the data set to be predicted of the specified service by adopting a service prediction model aiming at the specified service to obtain a prediction value of each data to be predicted; the service prediction model is formed by combining a binary classification model and a regression model;

D2. The apparatus of D1, the prediction unit comprising:

the first prediction module is used for predicting the data to be predicted by adopting the two-classification model to obtain two-classification label values of the data to be predicted;

the second prediction module is used for determining the predicted value of each data to be predicted with the second classification label value as the second classification label value;

and the third prediction module is used for predicting the data to be predicted with the two-classification label value as the first two-classification label value by adopting the regression model to obtain the predicted value of the data to be predicted with the two-classification label value as the first two-classification label value.

E1. A computer-readable storage medium, which includes a stored program, and in which, when the program runs, a device in which the storage medium is located is controlled to execute a training method of a traffic prediction model according to any one of items A1 to a14, or execute a traffic prediction method according to any one of items B1 to B2.

F1. A storage management device, the storage management device comprising:

a memory for storing a program;

a processor, coupled to the memory, for executing the program to perform the training method of the traffic prediction model according to any one of items A1 to a14 or perform the traffic prediction method according to any one of items B1 to B2.

It will be appreciated that the relevant features of the method and apparatus described above are referred to one another. In addition, "first", "second", and the like in the above embodiments are for distinguishing the embodiments, and do not represent merits of the embodiments.

It is clear to those skilled in the art that, for convenience and brevity of description, the specific working processes of the above-described systems, apparatuses and units may refer to the corresponding processes in the foregoing method embodiments, and are not described herein again.

The algorithms and displays presented herein are not inherently related to any particular computer, virtual machine, or other apparatus. Various general purpose systems may also be used with the teachings herein. The required structure for constructing such a system will be apparent from the description above. Moreover, the present invention is not directed to any particular programming language. It is appreciated that a variety of programming languages may be used to implement the teachings of the present invention as described herein, and any descriptions of specific languages are provided above to disclose the best mode of the invention.

In the description provided herein, numerous specific details are set forth. It is understood, however, that embodiments of the invention may be practiced without these specific details. In some instances, well-known methods, structures and techniques have not been shown in detail in order not to obscure an understanding of this description.

Similarly, it should be appreciated that in the foregoing description of exemplary embodiments of the invention, various features of the invention are sometimes grouped together in a single embodiment, figure, or description thereof for the purpose of streamlining the disclosure and aiding in the understanding of one or more of the various inventive aspects. However, the disclosed method should not be interpreted as reflecting an intention that: rather, the invention as claimed requires more features than are expressly recited in each claim. Rather, as the following claims reflect, inventive aspects lie in less than all features of a single foregoing disclosed embodiment. Thus, the claims following the detailed description are hereby expressly incorporated into this detailed description, with each claim standing on its own as a separate embodiment of this invention.

Those skilled in the art will appreciate that the modules in the device in an embodiment may be adaptively changed and disposed in one or more devices different from the embodiment. The modules or units or components in the embodiments may be combined into one module or unit or component, and furthermore, may be divided into a plurality of sub-modules or sub-units or sub-components. All of the features disclosed in this specification (including any accompanying claims, abstract and drawings), and all of the processes or elements of any method or apparatus so disclosed, may be combined in any combination, except combinations where at least some of such features and/or processes or elements are mutually exclusive. Each feature disclosed in this specification (including any accompanying claims, abstract and drawings) may be replaced by alternative features serving the same, equivalent or similar purpose, unless expressly stated otherwise.

Furthermore, those skilled in the art will appreciate that while some embodiments described herein include some features included in other embodiments, rather than other features, combinations of features of different embodiments are meant to be within the scope of the invention and form different embodiments. For example, in the following claims, any of the claimed embodiments may be used in any combination.

The various component embodiments of the invention may be implemented in hardware, or in software modules running on one or more processors, or in a combination thereof. Those skilled in the art will appreciate that a microprocessor or Digital Signal Processor (DSP) may be used in practice to implement some or all of the functions of some or all of the components of the method, apparatus and framework for operation of a deep neural network model in accordance with embodiments of the present invention. The present invention may also be embodied as apparatus or device programs (e.g., computer programs and computer program products) for performing a portion or all of the methods described herein. Such programs implementing the present invention may be stored on computer-readable media or may be in the form of one or more signals. Such a signal may be downloaded from an internet website or provided on a carrier signal or in any other form.

It should be noted that the above-mentioned embodiments illustrate rather than limit the invention, and that those skilled in the art will be able to design alternative embodiments without departing from the scope of the appended claims. In the claims, any reference signs placed between parentheses shall not be construed as limiting the claim. The word "comprising" does not exclude the presence of elements or steps not listed in a claim. The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements. The invention may be implemented by means of hardware comprising several distinct elements, and by means of a suitably programmed computer. In the unit claims enumerating several means, several of these means may be embodied by one and the same item of hardware. The usage of the words first, second and third, etcetera do not indicate any ordering. These words may be interpreted as names.

Claims

1. A method for training a business prediction model is characterized by comprising the following steps:

acquiring a training sample data set of a specified service, wherein the training sample data set comprises a plurality of training sample data, and each training sample data has an original label value thereof; wherein the specified service is at least any one of the following: predicting sales volume of retail goods, predicting overdue parameter values or income parameter values of bank lending products and predicting income parameter values of financial products;

determining a two-class label value of each training sample data based on an original label value and a preset label threshold value of each training sample data, wherein the two-class label value of the training sample data of which the original label value is equal to or greater than the preset label threshold value is determined as a first two-class label value, and the first two-class label value is a preset first threshold value; determining the two classification label values of the training sample data of which the original label values are smaller than a preset label threshold value in the training sample data set as second two classification label values, wherein the second two classification label values are preset second threshold values, and the second threshold values are different from the first threshold values;

performing two-classification training based on each training sample data in the training sample data set and the corresponding two-classification label values thereof to obtain two-classification models; performing regression training based on the training sample data with the first secondary classification label value in the training sample data set and the original label value corresponding to the training sample data to obtain a regression model;

2. The method of claim 1, wherein the original label value of each training sample data is a parameter value to be predicted in the specified service;

when the appointed service is used for predicting overdue parameter values of bank lending products, the original label values of the training sample data are the overdue parameter values of the bank lending products, wherein the overdue parameter values of the bank lending products are at least any one of the following: amount and rate of overdue;

3. The method of claim 1, wherein the combining the two classification models and the regression model to obtain a traffic prediction model comprises:

testing the binary classification model and the regression model by using each test sample data in the test sample data set of the specified service;

4. The method of claim 3, wherein the testing the binary model and the regression model using each test sample data in the test sample data set for the specified service comprises:

determining at least one comparative prediction index based on the original label value of each test sample data in the test sample data set;

determining whether the binary model and the regression model pass the test based on the at least one comparison predictor and the model predictor.

5. The method of claim 4, wherein said determining a predicted value for each of said test pattern data using said two classification models and said regression model comprises:

predicting each test sample data by adopting the two classification models to obtain a two classification label value of each test sample data;

and predicting each test sample data with the first second-class label value as the two-class label value by adopting the regression model to obtain the predicted value of each test sample data with the first second-class label value as the two-class label value.

6. The method of claim 4, wherein determining a model predictor from the predicted value of each of the test sample data comprises:

7. The method of claim 4, wherein determining at least one comparative predictor based on an original label value of each test sample data in the set of test sample data comprises:

8. The method of claim 4, wherein determining whether the bi-classification model and the regression model test pass based on the at least one comparison predictor and the model predictor comprises:

9. The method of claim 4, wherein determining whether the bi-classification model and the regression model test pass based on the at least one comparison predictor and the model predictor comprises:

the first formula is:

wherein, T is _i Representing an error index of the model prediction index and the ith comparison prediction index; the A represents the model predictor; b is described _i Characterizing the ith comparison prediction index;

10. The method according to any of claims 1-9, wherein said performing two-class training based on each training sample data in said set of training sample data and its respective corresponding two-class label value comprises:

and performing two-classification training on each training sample data in the training sample data set and the corresponding two-classification label value thereof by using the selected two-classification training method.

11. The method of claim 10,

when the number of continuous variables in each training sample data is larger than a preset continuous variable threshold, the selected binary training method is a gradient boosting decision tree GBDT method;

when the total data amount of each training sample data is larger than a preset data amount threshold value, the selected two-classification training method is a neural network method.

12. The method according to any of claims 1-9, wherein said performing regression training based on each training sample data having a first secondary label value in said set of training sample data and its respective corresponding original label value comprises:

13. The method of claim 12,

when the number of continuous variables in each training sample data with the first secondary classification label value is larger than a preset continuous variable threshold value, the selected regression training method is a gradient regression tree GBRT method;

14. A traffic prediction method, comprising:

predicting each data to be predicted in a data set to be predicted of a specified service by adopting a service prediction model for the specified service trained based on any one method of claims 1 to 13 to obtain a predicted value of each data to be predicted; the service prediction model is formed by combining a two-classification model and a regression model;

15. The method according to claim 14, wherein the predicting, by using a service prediction model for a specific service, each data to be predicted in the data set to be predicted of the specific service to obtain a predicted value of each data to be predicted comprises:

determining the predicted value of each data to be predicted with the second classification label value as the second classification label value;

16. An apparatus for training a traffic prediction model, comprising:

the system comprises an acquisition unit, a processing unit and a processing unit, wherein the acquisition unit is used for acquiring a training sample data set of a specified service, the training sample data set comprises a plurality of training sample data, and each training sample data has an original label value thereof; wherein, the specified service is at least any one of the following: predicting sales volume of retail goods, predicting overdue parameter values or income parameter values of bank lending products and predicting income parameter values of financial products;

the device comprises a determining unit, a judging unit and a judging unit, wherein the determining unit is used for determining two classification label values of each training sample data based on an original label value and a preset label threshold of each training sample data, and comprises the step of determining the two classification label values of the training sample data of which the original label value is equal to or greater than the preset label threshold as a first two classification label value, wherein the first two classification label value is a preset first threshold; determining the two classification label values of the training sample data of which the original label values are smaller than a preset label threshold value in the training sample data set as second two classification label values, wherein the second two classification label values are preset second threshold values, and the second threshold values are different from the first threshold values;

17. The apparatus according to claim 16, wherein the original label value of each training sample data is a parameter value to be predicted in the specified service;

18. The apparatus of claim 16, wherein the combining unit comprises:

19. The apparatus of claim 18, wherein the test module comprises:

the second determining submodule is used for determining at least one comparison prediction index based on the original label value of each test sample data in the test sample data set; determining whether the binary model and the regression model pass the test based on the at least one comparison predictor and the model predictor.

20. The apparatus according to claim 19, wherein the first determining sub-module is configured to predict each test sample data using the two classification models to obtain two classification label values of each test sample data; determining the predicted value of each test sample data with the second classification label value as the corresponding second classification label value; and predicting each test sample data with the first second-class label value as the two-class label value by adopting the regression model to obtain the predicted value of each test sample data with the first second-class label value as the two-class label value.

21. The apparatus of claim 19, wherein the first determining sub-module is configured to determine a mean absolute error or a root mean square error for each test sample data according to the predicted value of each test sample data; and determining the average absolute error or the root mean square error of each test sample data as the model prediction index.

22. The apparatus of claim 19, wherein the second determining sub-module is configured to determine a mean of the original label values of each test datum as a comparison predictor and/or determine a mode of the original label values of each test datum as a comparison predictor.

23. The apparatus of claim 19, wherein the second determining sub-module is configured to determine that the classification model and the regression model pass the test if the difference between the model predictor and each of the comparison predictors is within a predetermined range of difference corresponding to each of the comparison predictors.

24. The apparatus of claim 19, wherein the second determining sub-module is configured to calculate an error indicator for the model predictor and each of the comparison predictors using a first formula based on the at least one comparison predictor and the model predictor; if the error indexes of the model prediction indexes and the comparative prediction indexes are both larger than the preset error index threshold corresponding to the comparative prediction indexes, determining that the binary model and the regression model pass the test;

the first formula is:

wherein, T is _i Error index for representing model prediction index and ith comparison prediction indexMarking; the A represents the model predictor; b is _i And characterizing the ith comparison predictor.

25. The apparatus according to any one of claims 16-24, wherein the training unit comprises:

the training system comprises a first training module, a second training module and a third training module, wherein the first training module is used for selecting a two-classification training method based on two-classification training requirements and/or data characteristics of each training sample data, and the two-classification training requirements comprise training efficiency and/or training precision; the data characteristics include at least one of: the total data amount, the number of continuous variables and the number of discrete variables of each training sample data; and performing two-classification training on each training sample data in the training sample data set and the corresponding two-classification label value thereof by using the selected two-classification training method.

26. The apparatus of claim 25,

27. The apparatus according to any one of claims 16-24, wherein the training unit comprises:

the second training module is used for selecting a regression training method based on regression training requirements and/or data characteristics of training sample data with first secondary classification label values, wherein the regression training requirements comprise training efficiency and/or training precision; the data characteristics include at least one of: the total data amount, the number of continuous variables and the number of discrete variables of each training sample data; and performing regression training on each training sample data with the first secondary classification label value in the training sample data set and the original label value corresponding to each training sample data by using the selected regression training method.

28. The apparatus of claim 27,

29. A traffic prediction apparatus, comprising:

a prediction unit, configured to predict, by using a service prediction model for a specific service trained by an apparatus according to any one of claims 16 to 28, each piece of data to be predicted in a data set to be predicted of the specific service, so as to obtain a prediction value of each piece of data to be predicted; the service prediction model is formed by combining a binary classification model and a regression model;

30. The apparatus of claim 29, wherein the prediction unit comprises:

31. A computer-readable storage medium, comprising a stored program, wherein the program when executed controls an apparatus on which the storage medium is located to perform the method for training a traffic prediction model according to any one of claims 1 to 13, or to perform the method for traffic prediction according to any one of claims 14 to 15.

32. A storage management apparatus, characterized in that the storage management apparatus comprises:

a memory for storing a program;

a processor, coupled to the memory, for executing the program to perform a method of training a traffic prediction model as claimed in any one of claims 1 to 13, or to perform a method of traffic prediction as claimed in any one of claims 14 to 15.