Risk prediction method and device, electronic equipment and storage medium

Abstract

The embodiment of the invention discloses a risk prediction method, device and equipment. The method comprises the following steps: acquiring risk feature data of a target user; respectively inputting the risk feature data into at least two risk prediction models, respectively predicting risk occurrence probabilities of the target user on a plurality of risk categories by each risk prediction model based on the risk feature data, and outputting risk probability distribution information; wherein the at least two risk prediction models are obtained by training classifiers of different types through a risk prediction data set, and each risk prediction model is obtained by training a classifier of one type; performing information fusion based on the D-S evidence theory on the risk probability distribution information output by each risk prediction model to obtain fusion probability distribution information corresponding to the target user; and predicting the risk category of the target user based on the fusion probability distribution information corresponding to the target user.

Description

Technical Field

[0001] This document relates to the field of computer technology, and in particular to a risk prediction method, device, electronic device and storage medium. Background Art

[0002] Risk prediction is a key step in risk management for financial institutions such as banks. By predicting the risk of users, the risk category of users can be determined, and effective risk management can be carried out for users according to their risk categories. In the early days, risk prediction for users was usually carried out through linear discriminant, logistic regression, etc. Now, with the rapid development of technologies such as big data and machine learning, risk prediction for users is gradually carried out through machine learning methods. When predicting the risk of users through machine learning methods, classifiers are usually trained based on risk prediction data sets to obtain risk prediction models, and risk prediction for users is carried out based on the risk prediction model. In order to obtain a higher risk prediction accuracy, this method requires a large number of parameters to be set and adjusted when training the risk prediction model, and its training process is complicated. Summary of the invention

[0003] The purpose of the embodiments of this specification is to provide a risk prediction method, device, electronic device and storage medium, which are used to reduce the training complexity of the risk prediction model while obtaining higher risk prediction accuracy.

[0004] In a first aspect, an embodiment of the present specification provides a risk prediction method, the method comprising: Obtain risk characteristic data of target users; The risk feature data are respectively input into at least two risk prediction models, and each risk prediction model predicts the probability of risk occurrence of the target user in multiple risk categories based on the risk feature data, and outputs risk probability distribution information; wherein the risk probability distribution information includes: the probability of risk occurrence of the target user in the multiple risk categories, and the at least two risk prediction models are obtained by training classifiers of different types using risk prediction data sets, and each risk prediction model is obtained by training a classifier of one type; Performing information fusion based on DS evidence theory on the risk probability distribution information output by each risk prediction model to obtain fused probability distribution information corresponding to the target user; Based on the fused probability distribution information corresponding to the target user, the risk category of the target user is predicted.

[0005] In a second aspect, an embodiment of the present specification provides a risk prediction device, the device comprising: A feature acquisition module is used to obtain risk feature data of target users; A probability prediction module, used for inputting the risk feature data into at least two risk prediction models respectively, and each risk prediction model predicts the probability of risk occurrence of the target user in multiple risk categories based on the risk feature data respectively, and outputs risk probability distribution information; wherein the risk probability distribution information includes: the probability of risk occurrence of the target user in the multiple risk categories, the at least two risk prediction models are obtained by training classifiers of different types using the risk prediction data set, and each risk prediction model is obtained by training a classifier of one type; A probability fusion module is used to perform information fusion on the risk probability distribution information output by each risk prediction model based on DS evidence theory to obtain fused probability distribution information corresponding to the target user; The category prediction module is used to predict the risk category of the target user based on the fused probability distribution information corresponding to the target user.

[0006] In a third aspect, an embodiment of the present specification provides an electronic device, comprising a processor and a memory, wherein the memory stores programs or instructions that can be executed on the processor, and when the program or instructions are executed by the processor, the steps of the risk prediction method described in the first aspect of the embodiment of the present specification are implemented.

[0007] In a fourth aspect, an embodiment of the present specification provides a storage medium having computer instructions stored thereon, which, when executed by a processor, implement the steps of the risk prediction method as described in the first aspect of the embodiment of the present specification.

[0008] The risk prediction method, device, electronic device and storage medium provided in the embodiments of this specification obtain at least two risk prediction models based on training of classifiers of different types, input the risk feature data of the target user into at least two risk prediction models respectively, and each risk prediction model predicts the probability of risk occurrence of the target user in multiple risk categories based on the risk feature data respectively, and outputs risk probability distribution information, and then performs information fusion based on DS evidence theory on the risk probability distribution information output by each risk prediction model to obtain fused probability distribution information, and predicts the risk category of the target user based on the fused probability distribution information, which can utilize the complementarity of different classifiers to improve the accuracy of determining the user's risk category and improve the versatility and robustness of risk prediction. In addition, this method allows the training process of the risk prediction model to be lightweight, and only a small number of parameters are allowed to be set or adjusted during training, thereby reducing the training complexity of the risk prediction model while improving the accuracy of risk prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the embodiments of this specification or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this specification. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0010] Figure 1 A schematic diagram of an application scenario of the risk prediction method provided in some embodiments of this specification; Figure 2 A schematic diagram of a risk prediction method provided for some embodiments of this specification; Figure 3 A schematic diagram of another application scenario of the risk prediction method provided in some embodiments of this specification; Figure 4 A schematic diagram of the information fusion process provided for some embodiments of this specification; Figure 5 A schematic diagram of the training process of the risk prediction model provided for some embodiments of this specification; Figure 6 A schematic diagram of a target dimension determination process provided for some embodiments of this specification; Figure 7 Training examples of risk prediction models provided for some embodiments of this specification; Figure 8 A schematic diagram of a module of a risk prediction device provided in some embodiments of this specification; Fig. 9 A schematic diagram of the structure of an electronic device provided for some embodiments of this specification. DETAILED DESCRIPTION

[0011] The embodiments of this specification provide a risk prediction method, device and electronic device.

[0012] In order to enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below in conjunction with the drawings in the embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this specification, not all of the embodiments. Based on the embodiments in this specification, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of this specification.

[0013] See also Figure 1 , is a schematic diagram of an application scenario of the risk prediction method provided in some embodiments of this specification. Figure 1As shown, in this application scenario, two risk prediction models are provided, each of which is used to predict the risk probability of the target user in M ​​risk categories based on the risk characteristic data of the target user, and output the risk probability distribution information ,in, = , Represents the probability of occurrence of the risk of the jth risk category predicted by the ith risk prediction model. When it is necessary to perform risk prediction on the target user, the risk characteristic data of the target user can be obtained, and the risk characteristic data can be input into the first risk prediction model and the second risk prediction model respectively. After receiving the input risk characteristic data, the first risk prediction model can perform risk prediction on the target user based on the risk characteristic data, determine the probability of occurrence of the risk of the target user in risk category 1, risk category 2, risk category 3, ... risk category M, and output the first risk probability distribution information. ,in, = After receiving the input risk feature data, the second risk prediction model can perform risk prediction on the target user based on the risk feature data, determine the probability of the target user's risk occurrence in risk category 1, risk category 2, risk category 3, ..., risk category M, and output the second risk probability distribution information. ,in, = After obtaining the first risk probability distribution information output by the first risk prediction model and the second risk probability distribution information output by the second risk prediction model After that, the first risk probability distribution information can be and the second risk probability distribution information Perform information fusion based on DS evidence theory to obtain fusion probability distribution information , = , then, the fusion probability distribution information can be The fusion probability distribution information corresponding to the target user is determined, and the risk category with the largest fusion probability in the fusion probability information corresponding to the target user is predicted as the risk category of the target user.

[0014] Among them, the risk can be the user's personal financial risk, such as the user's fraud risk and credit risk, and the above-mentioned risk category can be the risk category of the user's financial risk, such as the risk category of fraud risk and the risk category of credit risk. The risk prediction method provided in the embodiment of this specification can be used in the scenario where communication operators and financial institutions conduct joint risk prediction. The risk characteristic data of the target user can be obtained based on the communication side user data of the communication operator, and the final predicted risk category of the target user can be provided to the financial institution, such as, sent to the financial institution system of the financial institution for downstream processing by the financial institution system. In combination with the above-mentioned application scenarios, the risk prediction method provided in the embodiment of this specification is now described in detail.

[0015] Figure 2 Schematic diagram of the risk prediction method provided by some embodiments of this specification. Figure 2 As shown, in some embodiments of this specification, the risk prediction method includes the following steps: S202, obtaining risk characteristic data of the target user.

[0016] The target user is a user for whom target risk prediction is to be performed, and the risk characteristic data is user characteristic data of the target user in a target dimension, and the target dimension is a user characteristic dimension required for performing the target risk prediction. When performing target risk prediction, the characteristic data of the target user in the target dimension of the target risk prediction can be obtained to obtain the risk characteristic data of the target user.

[0017] For example, in the target risk prediction for the first financial risk prediction conducted jointly by the communication operator and the financial institution, the risk characteristic data is the characteristic data used for the first financial risk prediction. The user characteristic dimensions required for the first financial risk prediction may be the first dimension and the second dimension. When making the first financial risk prediction, the characteristic data of the target user in the first dimension and the second dimension may be extracted from the communication side user data of the communication operator, and the risk characteristic data of the target user may be obtained based on the characteristic data of the target user in the first dimension and the second dimension. Among them, the first risk is, for example, the risk of user fraud, the first dimension is, for example, the call duration dimension, and the second dimension is, for example, the call object dimension.

[0018] For another example, in the target risk prediction for the second financial risk prediction conducted jointly by the communication operator and the financial institution, the risk characteristic data is the characteristic data used for the second financial risk prediction. The user characteristic dimensions required for the second financial risk prediction may be the first dimension, the second dimension, and the third dimension. When performing the second risk prediction, the characteristic data of the target user in the first dimension, the second dimension, and the third dimension may be extracted from the user data on the communication side of the communication operator, and the risk characteristic data of the target user may be obtained based on the characteristic data of the target user in the first dimension, the second dimension, and the third dimension. Among them, the second risk may be, for example, the user credit risk, the first dimension may be, for example, the communication behavior dimension, the second dimension may be, for example, the consumption behavior dimension, and the third dimension may be, for example, the location dimension.

[0019] In the above example, the risk characteristic data of the target user is obtained based on the communication side user data provided by the communication operator. The communication side user data provided by the communication operator is the user data on the communication operator side, which contains a large amount of user information of the communication operator, such as the user's daily location record information, visited web page information, mobile terminal attribute information, communication behavior information, consumption behavior information, etc. By extracting the user characteristic data of the characteristic dimensions required for financial risk prediction based on the communication side user data of the communication operator, and obtaining the risk characteristic data for financial risk prediction based on the user characteristic data of the required characteristic dimensions, the user data on the communication operator side can be used to empower the financial risk prediction on the financial institution side, thereby improving the accuracy of the user's financial risk prediction.

[0020] It is understandable that in the above example, the risk characteristic data of the target user is obtained only based on the communication side user data provided by the communication operator. In some risk predictions, the required user characteristic dimensions need to be obtained based on the communication side user data of the communication operator and the financial side user data of the financial institution. In these risk predictions, when obtaining the risk characteristic data of the target user, the user characteristic data of the target user in the characteristic dimensions required for the risk prediction can be extracted from the mixed user data of the communication side user data provided by the communication operator and the financial side user data provided by the financial institution, thereby obtaining the risk characteristic data of the target user.

[0021] For example, assuming that the target risk prediction is a joint third financial risk prediction by a communication operator and a financial institution, the feature dimensions required for the third risk prediction may be the first dimension and the second dimension on the communication side and the third dimension on the financial institution side. When performing the third risk prediction, the feature information of the target user in the feature dimensions required for the third risk prediction may be extracted from the mixed user data of the communication side user data provided by the communication operator and the financial institution side user data provided by the financial institution, thereby obtaining the risk feature data of the target user. Among them, the third risk may be, for example, the user credit risk, the first dimension may be, for example, the terminal attribute dimension, the second dimension may be, for example, the location attribute dimension, and the third dimension may be, for example, the asset dimension.

[0022] S204, inputting the risk characteristic data into at least two risk prediction models respectively, and each risk prediction model predicts the probability of risk occurrence of the target user in multiple risk categories based on the risk characteristic data respectively, and outputs risk probability distribution information.

[0023] The risk probability distribution information includes: the risk probability of the target user in the multiple risk categories, that is, the risk probability of the target user in each risk category of the multiple risk categories, which is, for example, an array composed of the risk probability of the target user in each risk category. Each risk prediction model outputs a risk probability distribution information. The at least two risk prediction models are obtained by training different types of classifiers using the same risk prediction data set, and each risk prediction model is trained based on a type of classifier.

[0024] For example, Figure 1 As shown, the risk characteristic data of the target user can be input into the first risk prediction model and the second risk prediction model respectively, and the first risk prediction model predicts the probability of the target user's risk occurrence in risk category 1, risk category 2, risk category 3, ..., risk category M based on the risk characteristic data, and outputs the risk probability distribution information The second risk prediction model predicts the probability of risk occurrence of the target user in risk category 1, risk category 2, risk category 3, ..., risk category M based on the risk feature data, and outputs risk probability distribution information .

[0025] The first risk prediction model can be obtained by training classifier 1 based on the risk prediction data set, and the second risk prediction model can be obtained by training classifier 2 based on the risk prediction data set, and classifier 1 and classifier 2 are of different types. For example, classifier 1 can be a DT (Decision Tree) classifier, and classifier 2 can be a RF (Random Forests) classifier.

[0026] Of course, classifier 1 and classifier 2 may also be other types of classifiers. For example, classifier 1 and classifier 2 may be any two of the following types of classifiers: DT (Decision Tree) classifier, RF (Random Forests) classifier, SVM (Support Vector Machine) classifier, KNN (K-Nearest Neighbors) classifier, BPNN (Back Propagation Neural Network) classifier, XGboost (eXtreme Gradient Boosting) classifier.

[0027] Among them, the risk prediction data set is a training data set used to train a risk prediction model for predicting the target risk.

[0028] For example, in a scenario where the target risk prediction is a first financial risk prediction conducted jointly by a telecommunications operator and a financial institution, the risk prediction set is a data set used to train a risk prediction model for the first financial risk prediction. In a scenario where the target risk prediction is a second financial risk prediction conducted jointly by a telecommunications operator and a financial institution, the risk prediction set is a data set used to train a risk prediction model for the second financial risk prediction.

[0029] S206, performing information fusion based on DS evidence theory on the risk probability distribution information output by each risk prediction model to obtain fused probability distribution information corresponding to the target user.

[0030] Among them, DS (Dempster-Shafer) evidence theory is a method for processing uncertain information, which can fuse multi-source information to obtain more complete, reliable and accurate information. In specific implementation, the risk probability distribution information output by each risk prediction model can be fused based on the combination rule of DS (Dempster-Shafer) evidence theory, that is, DS information fusion rule, to obtain fusion analysis probability distribution information.

[0031] For example, Figure 1 As shown, at least two risk prediction models are two risk prediction models, and the risk probability distribution information output by the first risk prediction model can be fused based on the first DS information fusion rule. The risk probability distribution information output by the second risk prediction model Perform information fusion to obtain first risk fusion probability distribution information, and convert the first fusion probability distribution information into , determine the fusion probability distribution information corresponding to the target user.

[0032] The first DS information fusion rule may be:

[0033] in,

[0034] in, The risk probability distribution information output by the first risk prediction model; The risk probability distribution information output by the second risk prediction model; Indicates the multiplication of the probability results output by the two models; Represents the set of interest, that is, the case where the outputs of the two models are consistent. By summing up, we can integrate the support of different sources of evidence and calculate the final joint trust value; Empty set Represents a contradictory set, that is, the situation where the outputs of the two models are inconsistent. All these conflicting terms are added together to obtain the total conflict coefficient ,When combining the classes, normalization can ensure that the fused trust assignment can reasonably reflect the results of the common support of the classifiers rather than being disturbed by conflicting evidence.

[0035] S208: Predict the risk category of the target user based on the fused probability distribution information corresponding to the target user.

[0036] Specifically, the risk category with the largest fusion probability value in the fusion probability distribution information may be predicted as the risk category of the target user.

[0037] For example, Figure 1 As shown, the fusion probability distribution information is , = ,like Maximum, that is, the fusion probability value of the second risk type is the largest, then the second risk type can be predicted as the risk category of the target user.

[0038] The risk prediction method provided in the embodiments of this specification obtains at least two risk prediction models based on training of classifiers of different types, inputs the risk feature data of the target user into at least two risk prediction models respectively, and each risk prediction model predicts the probability of risk occurrence of the target user in multiple risk categories based on the risk feature data respectively, and outputs risk probability distribution information, and then performs information fusion based on DS evidence theory on the risk probability distribution information output by each risk prediction model to obtain fused probability distribution information, and predicts the risk category of the target user based on the fused probability distribution information, which can utilize the complementarity of different classifiers to improve the accuracy of determining the user's risk category and improve the versatility and robustness of risk prediction. In addition, this method allows the training process of the risk prediction model to be lightweight, and only a small number of parameters are allowed to be set or adjusted during training, thereby reducing the training complexity of the risk prediction model while improving the accuracy of risk prediction.

[0039] In the above embodiment, the risk prediction method provided by the embodiment of this specification is described by taking at least two risk prediction models as two risk prediction models as an example. Furthermore, in order to improve the accuracy of the prediction of the risk category of the target user, in some embodiments of this specification, a larger number of risk prediction models can be obtained based on more types of classifier training, and the fused probability distribution information can be obtained based on the risk probability distribution information output by the larger number of risk prediction models, and then the risk category of the target user can be predicted based on the fused probability distribution information.

[0040] Figure 3 This is a schematic diagram of an application scenario of the risk prediction method provided in some embodiments of this specification. In this application scenario, N risk prediction models are provided. The N risk prediction models are obtained by training N different types of classifiers based on the same risk prediction data set, and each risk prediction model is obtained by training a classifier of one type. Figure 3 , the risk prediction method provided in the embodiments of this specification is further explained.

[0041] like Figure 3 As shown, when N risk prediction models are provided, the risk characteristic data of the target user can be respectively input into the first risk prediction model, the second risk prediction model, the third risk prediction model, ..., the Nth risk prediction model; the first risk prediction model predicts the risk probability of the target user in M ​​risk categories based on the risk characteristic data, and outputs the first risk probability distribution information ; The second risk prediction model predicts the risk occurrence probability of the target user in M ​​risk categories based on the risk feature data, and outputs the second risk probability distribution information ; The third risk prediction model predicts the risk occurrence probability of the target user in M ​​risk categories based on the risk feature data, and outputs the third risk probability distribution information ...; The Nth risk prediction model predicts the risk occurrence probability of the target user in M ​​risk categories based on the risk feature data, and outputs the Nth risk probability distribution information ; Afterwards, the N risk probability distribution information can be obtained , , ,…… Information fusion based on DS evidence theory is performed to obtain fused probability distribution information corresponding to the target user, and then the risk category of the target user is predicted based on the fused probability distribution information corresponding to the target user.

[0042] Wherein, N represents the number of risk prediction models, which is any integer greater than 2; M represents the number of risk categories, which is any integer greater than 2; represents the risk probability distribution information output by the ith risk prediction model, that is, the basic probability distribution of the ith risk prediction model on multiple risk categories, = , The j-th risk category of the i-th risk prediction model, j=1,2,3,…M.

[0043] In some embodiments, at least two risk prediction models are two risk prediction models, that is, N=2. In this case, as in the above embodiment, the risk probability distribution information output by the first risk prediction model can be fused based on the first DS information fusion rule. The risk probability distribution information output by the second risk prediction model Information fusion is performed, and the first fused probability distribution information obtained after the fusion is determined as the fused probability distribution information corresponding to the target user.

[0044] In some embodiments, at least two risk prediction models are more than two risk prediction models, that is, N is an integer greater than 3. In this case, the risk probability distribution information output by each risk prediction model can be subjected to multiple iterative information fusion based on DS evidence theory to obtain fused probability distribution information. Multiple iterative information fusion refers to the process of obtaining a fusion result through multiple iterations.

[0045] Figure 4 is a flowchart of information fusion provided by some embodiments of this specification. Figure 4As shown, when at least two risk prediction models are more than two risk prediction models, information fusion based on DS evidence theory is performed on the risk probability distribution information output by each risk prediction model, specifically including the following steps: S402, based on the first DS information fusion rule, fuse the risk probability distribution information output by the first risk prediction model with the risk probability distribution information output by the second risk prediction model to obtain the first fused probability distribution information, set i to 1, and go to step S404.

[0046] The first DS information fusion rule is a rule for performing DS information fusion on the risk probability distribution information output by the two risk prediction models, which is, for example:

[0047] in,

[0048] in, The risk probability distribution information output by the first risk prediction model; The risk probability distribution information output by the second risk prediction model; Indicates the multiplication of the probability results output by the two models; Represents the set of interest, that is, the case where the outputs of the two models are consistent. By summing up, we can integrate the support of different sources of evidence and calculate the final joint trust value; Empty set Represents a contradictory set, that is, the situation where the outputs of the two models are inconsistent. All these conflicting terms are added together to obtain the total conflict coefficient , when combining, through Normalization ensures that the ensemble's confidence assignments reasonably reflect the common support of the classifiers rather than being influenced by conflicting evidence.

[0049] Wherein, i is an iteration variable, 1<=i<=N-2, and N is the total number of risk prediction models.

[0050] S404: Based on the second DS information fusion rule, perform DS information fusion on the i-th fused probability distribution information and the (i+2)-th risk probability distribution information to obtain the (i+1)-th fused probability distribution information.

[0051] The second DS information fusion rule is to perform a DS information fusion rule on the fused probability distribution information obtained by fusing the risk probability distribution information output by the two risk prediction models and the risk probability distribution information output by another risk prediction model, which is, for example:

[0052] in, Fusion probability distribution information for the n-1th time; The risk probability distribution information output by the nth risk prediction model; Fusion probability distribution information for the n-2th time; Conflict coefficient. Represents the set of interest, that is, the case where the two outputs are consistent. By summing up, we can integrate the support of different evidence sources and calculate the final joint trust value; Empty set Represents a contradictory set, that is, the case where the two outputs are inconsistent. Adding all these conflicting terms together gives the total conflict coefficient , when combining, through Normalization ensures that the ensemble's confidence assignments reasonably reflect the common support of the classifiers rather than being influenced by conflicting evidence.

[0053] S406, increment i by 1 and go to the above step S404.

[0054] The above steps S404 to S406 are iteratively executed until i+2=N.

[0055] For example, Figure 3 As shown, when i=1, the first fusion probability distribution information can be fused based on the second DS information fusion rule. The risk probability distribution information output by the third risk prediction model Perform information fusion to obtain the second fusion probability distribution information When i=2, the second fusion probability distribution information can be fused based on the second DS information fusion rule. The risk probability distribution information output by the fourth risk prediction model Perform information fusion to obtain the third fusion probability distribution information When i=3, based on the second DS information fusion rule, the third fusion probability distribution information The risk probability distribution information output by the fifth risk prediction model . ...and so on, until the risk probability distribution information output by the last risk prediction model is traversed.

[0056] S408, determining the N-1th fusion probability distribution information obtained when i+2=N as the fusion probability distribution information corresponding to the target user.

[0057] It can be understood that when the risk probability distribution information output by the last risk prediction model is traversed, i+2=N, and this iteration is the last iteration. In this iteration, the N-2 fused probability distribution information is fused based on the second DS information fusion rule. The risk probability distribution information output by the Nth risk prediction model Perform information fusion and obtain the N-1th fusion probability distribution information The N-1th fusion probability distribution information obtained from the last iteration can be , determined as the fused probability distribution information corresponding to the target user. When predicting the risk category of the target user, the fused probability distribution information corresponding to the target user can be used to , predict the risk category of the target user. For example, the probability distribution information is integrated The risk category with the highest fusion probability is determined as the risk category of the target user.

[0058] When N is 2, the first fusion probability distribution information obtained in step S402 may be directly determined as the fusion probability distribution information corresponding to the target user.

[0059] The risk prediction method provided in the embodiments of this specification provides more risk prediction models, and more risk probability prediction models predict the probability of risk occurrence of the target user in multiple risk categories based on the risk feature data. The risk probability distribution information output by more risk prediction models is fused based on the DS evidence theory. This can combine the advantages of more classifiers, further improve the prediction accuracy, and further reduce the training complexity of the risk prediction model.

[0060] Among them, the more risk prediction model models are considered, the more models required for training are also needed. The appropriate number of risk prediction models can be determined by experiment. In the present specification embodiment, in order to achieve a balance between prediction accuracy and training complexity, the number of risk prediction models can be 6, and the 6 risk prediction models can be trained based on DT classifier, RF classifier, SVM classifier, KNN classifier, BPNN classifier, and XGboost classifier, respectively. Thus, it is possible to utilize the advantages of the above-mentioned appropriate number of classifiers to reduce training complexity while improving prediction accuracy.

[0061] Of course, it is understandable that when other number of risk prediction models are needed, other number of risk prediction models can also be obtained based on the above classifier training. Figure 3 In the application scenario shown, any N types of classifiers can be selected from the above 6 types of classifiers, and N risk prediction models can be trained based on the any N types of classifiers, where N is any number among 2, 3, 4, 5, and 6.

[0062] In the above embodiment, since multiple risk prediction models obtained by training based on classifiers of different types are set, the advantages of different classifiers are combined to improve the accuracy of prediction, and for each risk prediction model, it can allow a lower prediction accuracy. Therefore, when each type of classifier is trained to obtain a corresponding risk prediction model, the model parameters or hyperparameters that need to be set and adjusted can be less, and the training complexity will be lower. For example, when each type of classifier is trained to obtain a corresponding risk prediction model, the default hyperparameters of the training framework can be used directly without adjustment. In addition, setting and adjusting fewer model parameters or hyperparameters can also avoid overfitting of the risk prediction model and improve the generalization of the risk prediction model, thereby improving the versatility of the risk prediction method provided in the embodiments of this specification.

[0063] In the above embodiment, the risk characteristic data of the target user are respectively input into at least two risk prediction models, and the risk probability distribution information output by the at least two risk prediction models is subjected to information fusion based on DS information evidence theory, and the risk classifier of the target user is predicted based on the fused probability distribution information obtained by fusion, wherein at least two risk prediction models are obtained by training classifiers of different types based on the same risk prediction data set, thereby utilizing the complementarity of different types of classifiers to improve the accuracy and versatility of the prediction. Figure 5 This is a schematic diagram of the training process of the risk prediction model provided in some embodiments of this specification. Figure 5 , the risk prediction method provided in the embodiments of this specification is further explained.

[0064] like Figure 5 As shown, before the risk feature data are respectively input into at least two risk prediction models, the method provided in the embodiment of this specification further includes: S502, obtaining a risk prediction data set.

[0065] Among them, the risk prediction dataset is a dataset used to train the classifier to obtain a risk prediction model, which contains multiple sample data, each of which contains the risk characteristic data of a sample user and its corresponding risk category label. The risk prediction dataset varies with different application scenarios. In the joint risk prediction scenario of communication operators and financial institutions, the risk prediction sample dataset can be obtained based on the communication side user data of the communication operator and the financial side user data of the financial institution.

[0066] Among them, it can be understood that the data processing method during training and reasoning of the risk prediction model provided in the embodiments of this specification is consistent. Therefore, the risk characteristic data of the sample user in the risk prediction data set is similar to the risk characteristic data of the target user in the aforementioned embodiments. For example, if the risk characteristic data of the target user is the risk characteristic data used for target risk prediction, and multiple risk categories are risk categories for target risk prediction, then the risk prediction data set is a risk prediction data set for target risk prediction, and the risk characteristic data of the sample user is the risk characteristic data of the target risk of the sample user, which is the characteristic data of the target dimension required for target risk prediction in the user characteristic data of the sample user. The risk category label of the sample user is the label of the risk category of the target risk.

[0067] Among them, the target dimension required for target risk prediction can be determined according to the contribution of the user's characteristic data of each dimension to the risk category prediction of the target risk.

[0068] Figure 6 is a schematic diagram of a target dimension determination process provided in some embodiments of this specification. Figure 6 As shown, when a communication operator and a financial institution jointly conduct target risk prediction, a method for determining a target dimension of target risk prediction may include the following steps: S602: Acquire user characteristic data of sample users based on the communication-side user data of the communication operator.

[0069] In specific implementation, based on big data modeling technology, the communication-side user data of the communication operator can be mined and processed to obtain multiple user feature data, each user feature data corresponding to a sample user, and then the multiple user feature data are preprocessed. For example, user feature data with a missing value ratio exceeding a set threshold among the multiple user feature data are eliminated, and user feature data with outliers among the multiple user feature data are eliminated, so as to obtain a user feature data set.

[0070] S604: Obtain risk category labels of the sample users based on the financial user data of the financial institution.

[0071] Each sample user may correspond to a risk category label, and the risk category label is the category label of the target risk. For example, when the target risk is credit risk, the risk category label is the risk label of the credit risk, such as high credit risk, medium credit risk, low credit risk, etc.

[0072] In specific implementation, the financial side user data of financial institutions will contain risk label data of target risks, and these labels can be provided to the communication operator so that the communication operator can match these risk labels with the user feature data to obtain the risk label corresponding to each user feature data. For example, if user credit risk prediction is performed, the financial side user data will contain risk category labels of user credit risk, such as the user's credit rating label, and these labels can be provided to the communication operator for processing on the communication operator side.

[0073] S606: Based on the user feature data of the sample user and the risk category label of the sample user, select a dimension whose contribution to the risk category prediction of the target risk is within a contribution threshold range from multiple dimensions of the user feature data.

[0074] Among them, based on the user characteristic data of the sample users and the risk category label of the target risk of the sample users, the WOE (Weight of Evidence) value of each characteristic dimension of the user characteristic data can be calculated, the IV (Information Value) value can be calculated based on the WOE value of each dimension, and the dimension whose predicted contribution to the target risk is within the contribution threshold range is determined based on the IV value.

[0075] S608: Determine the target dimension based on the dimension within the contribution threshold range.

[0076] Among them, based on cross-validated recursive feature elimination, the dimensions whose contribution is within the contribution threshold range can be screened again, and the screened feature dimensions are determined as the target dimensions.

[0077] After determining the target dimension, the characteristic data of the target dimension in the user characteristic data can be determined as the risk characteristic data of the sample user; based on the risk characteristic data of the sample user and the risk category label of the target risk of the sample user, a risk prediction data set for predicting the target risk is obtained.

[0078] S504: Based on the risk prediction data set, at least two types of classifiers are trained separately to obtain at least two risk prediction models.

[0079] For example, Figure 7As shown, at least two types of classifiers are N types of classifiers, and N is any integer greater than or equal to 2. Based on the risk prediction data set, a first type of classifier can be trained to obtain a first risk prediction model; based on the risk prediction data set, a second type of classifier can be trained to obtain a second risk prediction model; based on the risk prediction data set, a third type of classifier can be trained to obtain a third risk prediction model, ... Based on the risk prediction data set, an Nth type of classifier can be trained to obtain an Nth risk prediction model.

[0080] Among them, the N types of classifiers can be any N types of the following classifiers: DT classifier, RF classifier, SVM classifier, KNN classifier, BPNN classifier, XGboost classifier.

[0081] For example, when there are 6 types of classifiers and 6 risk prediction models are trained, the first type of classifier may be a DT classifier, the second type of classifier may be an RF classifier, the third type of classifier may be an SVM classifier, the fourth type of classifier may be a KNN classifier, the fifth type of classifier may be a BPNN classifier, and the sixth type of classifier may be an XGboost classifier.

[0082] The risk prediction method provided by the embodiment of this specification is described above using the joint risk prediction scenario of a communication operator and a financial institution. The risk prediction method provided by the embodiment of this specification can also be used in other risk prediction scenarios, such as the risk prediction scenario of a communication operator alone, the risk prediction scenario of a financial institution alone, and the scenario of a communication operator and other institutions conducting joint risk prediction. The risk prediction methods for these scenarios are similar to the joint risk prediction scenario of a communication operator and a financial institution, except that the data provider based on which the risk characteristic data is obtained is different, or the object to which the final risk label is provided is different. For example, when a communication operator and other institutions other than a financial institution conduct a joint risk prediction, the risk characteristic data can be obtained from a mixture of user data on the communication side of the communication operator and user data on the other institution side of other institutions. The risk category of the target user predicted at the end can be provided to the other institution. For the risk prediction methods for these scenarios, reference can be made to the aforementioned embodiments, which will not be repeated here.

[0083] The above is a risk prediction method provided in the embodiment of this specification. Based on the same idea, the embodiment of this specification also provides a risk prediction device, such as Figure 8 shown.

[0084] The risk prediction device 800 includes: The feature acquisition module 801 is used to acquire risk feature data of the target user; The probability prediction module 802 is used to input the risk feature data into at least two risk prediction models respectively, and each risk prediction model predicts the risk occurrence probability of the target user in multiple risk categories based on the risk feature data, and outputs risk probability distribution information; wherein the risk probability distribution information includes: the risk occurrence probability of the target user in the multiple risk categories, and the at least two risk prediction models are obtained by training classifiers of different types based on the risk prediction data set, and each risk prediction model is obtained by training a classifier of one type; The probability fusion module 803 is used to perform information fusion on the risk probability distribution information output by each risk prediction model based on the DS evidence theory to obtain the fused probability distribution information corresponding to the target user; The category prediction module 804 is used to predict the risk category of the target user based on the fused probability distribution information corresponding to the target user.

[0085] In some embodiments of this specification, the at least two risk prediction models are two risk prediction models, and the probability fusion module 803 is specifically used to: Based on the first DS information fusion rule, information fusion is performed on the risk probability distribution information output by the first risk prediction model and the risk probability distribution information output by the second risk prediction model to obtain first fused probability distribution information; The first fusion probability distribution information is determined as the fusion probability distribution information corresponding to the target user.

[0086] In some embodiments of this specification, the at least two risk prediction models are more than two risk prediction models, and the probability fusion module 803 is specifically used to: Through the following processing, the risk probability distribution information output by each risk prediction model is subjected to multiple iterative information fusion based on DS evidence theory until i+2=N: Based on the second DS information fusion rule, the i-th fused probability distribution information is fused with the risk probability distribution information output by the i+2 risk prediction model to obtain the i+1-th fused probability distribution information; The N-1th fusion probability distribution information obtained when i+2=N is determined as the fusion probability distribution information corresponding to the target user; Wherein, N is the total number of the risk prediction models, i is the iterative variable, which increases by 1 after each iteration, 1<=i<=N-2, the initial value of i is 1, and the first fused probability distribution information is obtained by fusing the risk probability distribution information output by the first risk prediction model and the risk probability distribution information output by the second risk prediction model based on the first DS information fusion rule.

[0087] In some embodiments of this specification, the first DS information fusion rule includes:

[0088] in,

[0089] is the risk probability distribution information output by the first risk prediction model, The risk probability distribution information output by the second risk prediction model. Indicates the multiplication of the probability results output by the two models; Represents the set of interest; empty set Represents a contradictory set, Represents the conflict coefficient.

[0090] In some embodiments of this specification, the second DS information fusion rule includes:

[0091] in, Fusion probability distribution information for the n-1th time; The risk probability distribution information output by the nth risk prediction model; Fusion probability distribution information for the n-2th time; Conflict coefficient.

[0092] In some embodiments of the present specification, the risk characteristic data is risk characteristic data used for target risk prediction, the multiple risk categories are risk categories of target risks; the target risk prediction includes: joint financial risk prediction of communication operators and financial institutions; the acquisition of risk characteristic data of target users includes: Based on the communication-side user data of the communication operator, characteristic data of the target user in a target dimension is acquired, and based on the characteristic data of the target user in the target dimension, risk characteristic data of the target user is obtained; wherein the target dimension is a user characteristic dimension required for performing the target risk prediction; The risk prediction device further includes: The category sending module is used to send the predicted risk category of the target user to the financial institution system.

[0093] In some embodiments of the present specification, the device further comprises: A sample feature acquisition module, used to acquire user feature data of sample users based on communication side user data of communication operators; A sample category acquisition module, used to acquire a risk category label of the sample user based on the financial user data provided by the financial institution, where the risk category is the risk category of the target risk; A contribution determination module, configured to select, based on the user feature data of the sample user and the risk category label of the sample user, a dimension whose contribution to the risk category prediction of the target risk is within a contribution threshold range from the multiple dimensions of the user feature data; The target dimension determination module is used to determine the target dimension based on the dimension within the contribution threshold range.

[0094] In some embodiments of the present specification, the device further comprises: A sample risk feature acquisition module, used to determine the feature data of the target dimension in the user feature data as the risk feature data of the sample user; A risk prediction data set acquisition module, used to obtain a risk prediction data set based on the risk feature data of the sample users and the risk category labels of the sample users, wherein the risk prediction data set is a risk prediction data set for the target risk prediction; A training module is used to train at least two types of classifiers based on the risk prediction sample data set to obtain the at least two risk prediction models.

[0095] In some embodiments of the present specification, the at least two types of classifiers are at least two of the following types of classifiers: DT classifier, RF classifier, SVM classifier, SVM classifier, KNN classifier, BPNN classifier, XGBoost classifier.

[0096] In some embodiments of this specification, the category prediction module is specifically configured to include: Determine a risk category having a maximum fusion probability value in the fusion probability distribution information; The risk category with the largest fusion probability value is predicted as the risk category of the user.

[0097] The risk prediction device provided in the embodiment of this specification obtains at least two risk prediction models based on training of classifiers of different types, inputs the risk feature data of the target user into at least two risk prediction models respectively, and each risk prediction model predicts the probability of risk occurrence of the target user in multiple risk categories based on the risk feature data respectively, and outputs risk probability distribution information, and then performs information fusion based on DS evidence theory on the risk probability distribution information output by each risk prediction model to obtain fused probability distribution information, and predicts the risk category of the target user based on the fused probability distribution information, which can utilize the complementarity of different classifiers to improve the accuracy of determining the user's risk category and improve the versatility and robustness of risk prediction. In addition, this method allows the training process of the risk prediction model to be lightweight, and it allows only a small number of parameters to be set or adjusted during training, thereby reducing the training complexity of the risk prediction model while improving the accuracy of risk prediction.

[0098] It is understandable that the risk prediction device provided in the embodiments of this specification can achieve Figures 1 to 7 To avoid repetition, the various processes implemented in the risk prediction method embodiment will not be described here.

[0099] The above is a risk prediction device provided in the embodiment of this specification. Based on the same idea, the embodiment of this specification also provides an electronic device, such as Fig. 9 shown.

[0100] The electronic device may provide a terminal device or a server, etc. for the above-mentioned embodiments.

[0101] Electronic devices may have relatively large differences due to different configurations or performances, and may include one or more processors 901 and memory 902, and the memory 902 may store one or more storage applications or data. Among them, the memory 902 may be a short-term storage or a persistent storage. The application stored in the memory 902 may include one or more modules (not shown in the figure), and each module may include a series of computer executable instructions in the processing device of the model. Furthermore, the processor 801 may be configured to communicate with the memory 902 to execute a series of computer executable instructions in the memory 902 on the electronic device.

[0102] Specifically in this embodiment, the electronic device includes a memory and one or more programs, wherein the one or more programs are stored in the memory, and the one or more programs may include one or more modules, and each module may include a series of computer executable instructions in the processing device of the model, and the program or instruction is configured to implement the following process when executed by the processor: Obtain risk characteristic data of target users; The risk feature data are respectively input into at least two risk prediction models, and each risk prediction model predicts the probability of risk occurrence of the target user in multiple risk categories based on the risk feature data, and outputs risk probability distribution information; wherein the risk probability distribution information includes: the probability of risk occurrence of the target user in the multiple risk categories, and the at least two risk prediction models are obtained by training classifiers of different types using risk prediction data sets, and each risk prediction model is obtained by training a classifier of one type; Performing information fusion based on DS evidence theory on the risk probability distribution information output by each risk prediction model to obtain fused probability distribution information corresponding to the target user; Based on the fused probability distribution information corresponding to the target user, the risk category of the target user is predicted.

[0103] In some embodiments of the present specification, the at least two risk prediction models are two risk prediction models, and the information fusion based on DS evidence theory is performed on the risk probability distribution information output by each risk prediction model to obtain the fused probability distribution information corresponding to the target user, including: Based on the first DS information fusion rule, information fusion is performed on the risk probability distribution information output by the first risk prediction model and the risk probability distribution information output by the second risk prediction model to obtain first fused probability distribution information; The first fusion probability distribution information is determined as the fusion probability distribution information corresponding to the target user.

[0104] In some embodiments of the present specification, the at least two risk prediction models are more than two risk prediction models, and the risk probability distribution information output by each risk prediction model is subjected to information fusion based on DS evidence theory to obtain fused probability distribution information, including: Through the following processing, the risk probability distribution information output by each risk prediction model is subjected to multiple iterative information fusion based on DS evidence theory until i+2=N: Based on the second DS information fusion rule, the i-th fused probability distribution information is fused with the risk probability distribution information output by the i+2 risk prediction model to obtain the i+1-th fused probability distribution information; The N-1th fusion probability distribution information obtained when i+2=N is determined as the fusion probability distribution information corresponding to the target user; Wherein, N is the total number of the risk prediction models, i is the iterative variable, which increases by 1 after each iteration, 1<=i<=N-2, the initial value of i is 1, and the first fused probability distribution information is obtained by fusing the risk probability distribution information output by the first risk prediction model and the risk probability distribution information output by the second risk prediction model based on the first DS information fusion rule.

[0105] In some embodiments of this specification, the first DS information fusion rule includes:

[0106] in,

[0107] The risk probability distribution information output by the first risk prediction model; The risk probability distribution information output by the second risk prediction model; Indicates the multiplication of the probability results output by the two models; Represents the set of interest; empty set Represents a contradictory set, Represents the conflict coefficient.

[0108] In some embodiments of this specification, the second DS information fusion rule includes:

[0109] in, Fusion probability distribution information for the n-1th time; The risk probability distribution information output by the nth risk prediction model; Fusion probability distribution information for the n-2th time; Conflict coefficient.

[0110] In some embodiments of the present specification, the risk characteristic data is risk characteristic data used for target risk prediction, the multiple risk categories are risk categories of target risks; the target risk prediction includes: joint financial risk prediction of communication operators and financial institutions; the acquisition of risk characteristic data of target users includes: Based on the communication-side user data of the communication operator, characteristic data of the target user in a target dimension is acquired, and based on the characteristic data of the target user in the target dimension, risk characteristic data of the target user is obtained; wherein the target dimension is a user characteristic dimension required for performing the target risk prediction; After predicting the risk category of the target user based on the fused probability distribution information corresponding to the target user, the process further includes: The predicted risk category of the target user is sent to the financial institution system.

[0111] In some embodiments of the present specification, the method further includes: Based on the communication-side user data of the communication operator, user characteristic data of the sample users is obtained; Based on the financial user data provided by the financial institution, a risk category label of the sample user is obtained, where the risk category is the risk category of the target risk; Based on the user feature data of the sample user and the risk category label of the sample user, selecting a dimension whose contribution to the risk category prediction of the target risk is within a contribution threshold range from multiple dimensions of the user feature data; The target dimension is determined based on the dimension within the contribution threshold range.

[0112] In some embodiments of the present specification, before inputting the risk characteristic data into at least two risk prediction models respectively, the process further includes: Determining the characteristic data of the target dimension in the user characteristic data as the risk characteristic data of the sample user; Based on the risk feature data of the sample users and the risk category labels of the sample users, a risk prediction data set is obtained, where the risk prediction data set is a risk prediction data set for the target risk prediction; Based on the risk prediction sample data set, at least two types of classifiers are trained respectively to obtain the at least two risk prediction models.

[0113] In some embodiments of the present specification, the at least two types of classifiers are at least two of the following types of classifiers: DT classifier, RF classifier, SVM classifier, SVM classifier, KNN classifier, BPNN classifier, XGBoost classifier.

[0114] In some embodiments of the present specification, predicting the risk category of the target user based on the fused probability distribution information corresponding to the target user includes: Determine a risk category having a maximum fusion probability value in the fusion probability distribution information; The risk category with the largest fusion probability value is predicted as the risk category of the user.

[0115] The electronic device provided in the embodiment of this specification obtains at least two risk prediction models based on training of classifiers of different types, inputs the risk feature data of the target user into the at least two risk prediction models respectively, and each of the at least two risk prediction models predicts the probability of occurrence of the risk of the target user in multiple risk categories based on the risk feature data respectively, and outputs the risk probability distribution information, and then, performs information fusion based on the DS evidence theory on the risk probability distribution information output by each risk prediction model to obtain the fused probability distribution information, and predicts the risk category of the target user based on the fused probability distribution information, which can utilize the complementarity of different classifiers to improve the accuracy of determining the user's risk category and improve the versatility and robustness of risk prediction. In addition, this method allows the training process of the risk prediction model to be lightweight, and it allows only a small number of parameters to be set or adjusted during training, thereby reducing the training complexity of the risk prediction model while improving the accuracy of risk prediction.

[0116] In addition, corresponding to the risk prediction method provided in the above embodiments, based on the same technical concept, one or more embodiments of this specification further provide a storage medium for storing computer executable instructions. In a specific embodiment, the storage medium may be a USB flash drive, an optical disk, a hard disk, etc. When the computer executable instructions stored in the storage medium are executed by the processor, the following process can be implemented: Obtain risk characteristic data of target users; The risk feature data are respectively input into at least two risk prediction models, and each risk prediction model predicts the probability of risk occurrence of the target user in multiple risk categories based on the risk feature data, and outputs risk probability distribution information; wherein the risk probability distribution information includes: the probability of risk occurrence of the target user in the multiple risk categories, and the at least two risk prediction models are obtained by training classifiers of different types using risk prediction data sets, and each risk prediction model is obtained by training a classifier of one type; Performing information fusion based on DS evidence theory on the risk probability distribution information output by each risk prediction model to obtain fused probability distribution information corresponding to the target user; Based on the fused probability distribution information corresponding to the target user, the risk category of the target user is predicted.

[0117] In some embodiments of the present specification, the at least two risk prediction models are two risk prediction models, and the information fusion based on DS evidence theory is performed on the risk probability distribution information output by each risk prediction model to obtain the fused probability distribution information corresponding to the target user, including: Based on the first DS information fusion rule, information fusion is performed on the risk probability distribution information output by the first risk prediction model and the risk probability distribution information output by the second risk prediction model to obtain first fused probability distribution information; The first fusion probability distribution information is determined as the fusion probability distribution information corresponding to the target user.

[0118] In some embodiments of the present specification, the at least two risk prediction models are more than two risk prediction models, and the risk probability distribution information output by each risk prediction model is subjected to information fusion based on DS evidence theory to obtain fused probability distribution information, including: Through the following processing, the risk probability distribution information output by each risk prediction model is subjected to multiple iterative information fusion based on DS evidence theory until i+2=N: Based on the second DS information fusion rule, the i-th fused probability distribution information is fused with the risk probability distribution information output by the i+2 risk prediction model to obtain the i+1-th fused probability distribution information; The N-1th fusion probability distribution information obtained when i+2=N is determined as the fusion probability distribution information corresponding to the target user; Wherein, N is the total number of the risk prediction models, i is the iterative variable, which increases by 1 after each iteration, 1<=i<=N-2, the initial value of i is 1, and the first fused probability distribution information is obtained by fusing the risk probability distribution information output by the first risk prediction model and the risk probability distribution information output by the second risk prediction model based on the first DS information fusion rule.

[0119] In some embodiments of this specification, the first DS information fusion rule includes:

[0120] in,

[0121] is the risk probability distribution information output by the first risk prediction model, The risk probability distribution information output by the second risk prediction model. Indicates the multiplication of the probability results output by the two models; Represents the set of interest; empty set Represents a contradictory set, Represents the conflict coefficient.

[0122] In some embodiments of this specification, the second DS information fusion rule includes:

[0123] in, Fusion probability distribution information for the n-1th time; The risk probability distribution information output by the nth risk prediction model; Fusion probability distribution information for the n-2th time; Conflict coefficient.

[0124] In some embodiments of the present specification, the risk characteristic data is risk characteristic data used for target risk prediction, the multiple risk categories are risk categories of target risks; the target risk prediction includes: joint financial risk prediction of communication operators and financial institutions; the acquisition of risk characteristic data of target users includes: Based on the communication-side user data of the communication operator, characteristic data of the target user in a target dimension is acquired, and based on the characteristic data of the target user in the target dimension, risk characteristic data of the target user is obtained; wherein the target dimension is a user characteristic dimension required for performing the target risk prediction; After predicting the risk category of the target user based on the fused probability distribution information corresponding to the target user, the process further includes: The predicted risk category of the target user is sent to the financial institution system.

[0125] In some embodiments of the present specification, the method further includes: Based on the communication-side user data of the communication operator, user characteristic data of the sample users is obtained; Based on the financial user data provided by the financial institution, a risk category label of the sample user is obtained, where the risk category is the risk category of the target risk; Based on the user feature data of the sample user and the risk category label of the sample user, selecting a dimension whose contribution to the risk category prediction of the target risk is within a contribution threshold range from multiple dimensions of the user feature data; The target dimension is determined based on the dimension within the contribution threshold range.

[0126] In some embodiments of the present specification, before inputting the risk characteristic data into at least two risk prediction models respectively, the process further includes: Determining the characteristic data of the target dimension in the user characteristic data as the risk characteristic data of the sample user; Based on the risk feature data of the sample users and the risk category labels of the sample users, a risk prediction data set is obtained, where the risk prediction data set is a risk prediction data set for the target risk prediction; Based on the risk prediction sample data set, at least two types of classifiers are trained respectively to obtain the at least two risk prediction models.

[0127] In some embodiments of the present specification, the at least two types of classifiers are at least two of the following types of classifiers: DT classifier, RF classifier, SVM classifier, SVM classifier, KNN classifier, BPNN classifier, XGBoost classifier.

[0128] In some embodiments of the present specification, predicting the risk category of the target user based on the fused probability distribution information corresponding to the target user includes: Determine a risk category having a maximum fusion probability value in the fusion probability distribution information; The risk category with the largest fusion probability value is predicted as the risk category of the user.

[0129] The storage medium provided in the embodiment of this specification obtains at least two risk prediction models based on training of classifiers of different types, inputs the risk feature data of the target user into the at least two risk prediction models respectively, and each of the at least two risk prediction models predicts the probability of risk occurrence of the target user in multiple risk categories based on the risk feature data respectively, and outputs risk probability distribution information, and then, performs information fusion based on DS evidence theory on the risk probability distribution information output by each risk prediction model to obtain fused probability distribution information, and predicts the risk category of the target user based on the fused probability distribution information, which can utilize the complementarity of different classifiers to improve the accuracy of determining the user's risk category and improve the versatility and robustness of risk prediction. In addition, this method allows the training process of the risk prediction model to be lightweight, and it allows only a small number of parameters to be set or adjusted during training, thereby reducing the training complexity of the risk prediction model while improving the accuracy of risk prediction.

[0130] The above is a description of a specific embodiment of the present specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0131] In the 1990s, it was very clear whether the improvement of a technology was hardware improvement (for example, improvement of the circuit structure of diodes, transistors, switches, etc.) or software improvement (improvement of the method flow). However, with the development of technology, many improvements of the method flow today can be regarded as direct improvements of the hardware circuit structure. Designers almost always obtain the corresponding hardware circuit structure by programming the improved method flow into the hardware circuit. Therefore, it cannot be said that the improvement of a method flow cannot be implemented with a hardware entity module. For example, a programmable logic device (PLD) (such as a field programmable gate array (FPGA)) is such an integrated circuit whose logical function is determined by the user's programming of the device. Designers can "integrate" a digital system on a PLD by programming themselves, without having to ask chip manufacturers to design and make dedicated integrated circuit chips. Moreover, nowadays, instead of manually making integrated circuit chips, this kind of programming is mostly implemented by "logic compiler" software, which is similar to the software compiler used when developing and writing programs, and the original code before compilation must also be written in a specific programming language, which is called hardware description language (HDL). There is not only one kind of HDL, but many kinds, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, RHDL (Ruby Hardware Description Language), etc. The most commonly used ones are VHDL (Very-High-Speed ​​Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art should also know that it is only necessary to program the method flow slightly in the above-mentioned hardware description languages ​​and program it into the integrated circuit, and then it is easy to obtain the hardware circuit that implements the logic method flow.

[0132] The controller may be implemented in any suitable manner, for example, the controller may take the form of a microprocessor or processor and a computer-readable medium storing a computer-readable program code (e.g., software or firmware) executable by the (micro)processor, a logic gate, a switch, an application-specific integrated circuit (ASIC), a programmable logic controller, and an embedded microcontroller, examples of which include but are not limited to the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicone Labs C8051F320, and the memory controller may also be implemented as part of the control logic of the memory. It is also known to those skilled in the art that, in addition to implementing the controller in a purely computer-readable program code manner, the controller may be implemented in the form of a logic gate, a switch, an application-specific integrated circuit, a programmable logic controller, and an embedded microcontroller by logically programming the method steps. Therefore, such a controller may be considered as a hardware component, and the devices for implementing various functions included therein may also be considered as structures within the hardware component. Or even, the devices for implementing various functions may be considered as both software modules for implementing the method and structures within the hardware component.

[0133] The systems, devices, modules or units described in the above embodiments may be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, the computer may be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0134] For the convenience of description, the above devices are described in terms of functions and are divided into various units. Of course, when implementing one or more embodiments of this specification, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0135] Those skilled in the art will appreciate that the embodiments of this specification may be provided as methods, systems, or computer program products. Therefore, one or more embodiments of this specification may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, one or more embodiments of this specification may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0136] The embodiments of this specification are described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the embodiments of this specification. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable fraud case serial and parallel device to produce a machine, so that the instructions executed by the processor of the computer or other programmable fraud case serial and parallel device generate instructions for implementing the processes in the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0137] These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable fraud case serial and parallel device to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0138] These computer program instructions may also be loaded onto a computer or other programmable device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, whereby the instructions executed on the computer or other programmable device provide for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0139] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0140] Memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. Memory is an example of a computer-readable medium.

[0141] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined in this article, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0142] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0143] Those skilled in the art will appreciate that the embodiments of this specification may be provided as methods, systems or computer program products. Therefore, one or more embodiments of this specification may be in the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware. Furthermore, one or more embodiments of this specification may be in the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0144] One or more embodiments of the present specification may be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. One or more embodiments of the present specification may also be practiced in distributed computing environments where tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules may be located in local and remote computer storage media, including storage devices.

[0145] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0146] The above description is only an embodiment of this specification and is not intended to limit the present application. For those skilled in the art, this specification may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of this specification should be included in the scope of the claims of this specification.

Claims

1. A risk prediction method, characterized in that: The method comprises: Obtain risk characteristic data of target users; The risk feature data are respectively input into at least two risk prediction models, and each risk prediction model predicts the probability of risk occurrence of the target user in multiple risk categories based on the risk feature data, and outputs risk probability distribution information; wherein the risk probability distribution information includes: the probability of risk occurrence of the target user in the multiple risk categories, and the at least two risk prediction models are obtained by training classifiers of different types using risk prediction data sets, and each risk prediction model is obtained by training a classifier of one type; Performing information fusion based on DS evidence theory on the risk probability distribution information output by each risk prediction model to obtain fused probability distribution information corresponding to the target user; Based on the fused probability distribution information corresponding to the target user, the risk category of the target user is predicted.

2. The risk prediction method according to claim 1, characterized in that: The at least two risk prediction models are two risk prediction models, and the risk probability distribution information output by each risk prediction model is subjected to information fusion based on DS evidence theory to obtain fused probability distribution information corresponding to the target user, including: Based on the first DS information fusion rule, information fusion is performed on the risk probability distribution information output by the first risk prediction model and the risk probability distribution information output by the second risk prediction model to obtain first fused probability distribution information; The first fusion probability distribution information is determined as the fusion probability distribution information corresponding to the target user.

3. The risk prediction method according to claim 1, characterized in that: The at least two risk prediction models are more than two risk prediction models, and the risk probability distribution information output by each risk prediction model is subjected to information fusion based on DS evidence theory to obtain fused probability distribution information, including: Through the following processing, the risk probability distribution information output by each risk prediction model is subjected to multiple iterative information fusion based on DS evidence theory until i+2=N: Based on the second DS information fusion rule, the i-th fused probability distribution information is fused with the risk probability distribution information output by the i+2 risk prediction model to obtain the i+1-th fused probability distribution information; The N-1th fusion probability distribution information obtained when i+2=N is determined as the fusion probability distribution information corresponding to the target user; Wherein, N is the total number of the risk prediction models, i is the iterative variable, which increases by 1 after each iteration, 1<=i<=N-2, the initial value of i is 1, and the first fused probability distribution information is obtained by fusing the risk probability distribution information output by the first risk prediction model and the risk probability distribution information output by the second risk prediction model based on the first DS information fusion rule.

4. The risk prediction method according to claim 2 or 3, characterized in that: The first DS information fusion rule includes: in, The risk probability distribution information output by the first risk prediction model; The risk probability distribution information output by the second risk prediction model; Indicates the multiplication of the probability results output by the two models; Represents the set of interest; empty set Represents a contradictory set, Represents the conflict coefficient.

5. The risk prediction method according to claim 3, characterized in that: The second DS information fusion rule includes: in, Fusion probability distribution information for the n-1th time; The risk probability distribution information output by the nth risk prediction model; Fusion probability distribution information for the n-2th time; Conflict coefficient.

6. The risk prediction method according to claim 1, characterized in that: The risk characteristic data is risk characteristic data used for target risk prediction, and the multiple risk categories are risk categories of target risks; The target risk prediction includes: joint financial risk prediction of communication operators and financial institutions; the acquisition of risk characteristic data of target users includes: Based on the communication-side user data of the communication operator, characteristic data of the target user in a target dimension is acquired, and based on the characteristic data of the target user in the target dimension, risk characteristic data of the target user is obtained; wherein the target dimension is a user characteristic dimension required for performing the target risk prediction; After predicting the risk category of the target user based on the fused probability distribution information corresponding to the target user, the method further includes: The predicted risk category of the target user is sent to the financial institution system.

7. The risk prediction method according to claim 6, characterized in that: The method further comprises: Based on the communication-side user data of the communication operator, user characteristic data of the sample users is obtained; Based on the financial user data provided by the financial institution, a risk category label of the sample user is obtained, where the risk category is the risk category of the target risk; Based on the user feature data of the sample user and the risk category label of the sample user, selecting a dimension whose contribution to the risk category prediction of the target risk is within a contribution threshold range from multiple dimensions of the user feature data; The target dimension is determined based on the dimension within the contribution threshold range.

8. The risk prediction method according to claim 6, characterized in that: Before inputting the risk characteristic data into at least two risk prediction models respectively, the method further comprises: Determining the characteristic data of the target dimension in the user characteristic data as the risk characteristic data of the sample user; Based on the risk feature data of the sample users and the risk category labels of the sample users, a risk prediction data set is obtained, where the risk prediction data set is a risk prediction data set for the target risk prediction; Based on the risk prediction sample data set, at least two types of classifiers are trained respectively to obtain the at least two risk prediction models.

9. The risk prediction method according to claim 8, characterized in that: The at least two types of classifiers are at least two of the following types of classifiers: DT classifier, RF classifier, SVM classifier, SVM classifier, KNN classifier, BPNN classifier, XGBoost classifier.

10. The risk prediction method according to claim 1, characterized in that: The predicting the risk category of the target user based on the fused probability distribution information corresponding to the target user includes: Determine a risk category having a maximum fusion probability value in the fusion probability distribution information; The risk category with the largest fusion probability value is predicted as the risk category of the user.

Images