Risk prediction method and device, electronic equipment and storage medium

Abstract

The present disclosure relates to the technical field of data processing, and particularly relates to a risk prediction method and device, electronic equipment and a storage medium. The specific implementation scheme is: obtaining application information and electric verification information of a target user; obtaining a first risk assessment result based on the application information and the electric verification information by using a first risk assessment model; obtaining a second risk assessment result based on the application information and the electric verification information by using a second risk assessment model; and obtaining a third risk assessment result based on the first risk assessment result and the second risk assessment result. Through the first risk assessment model and the second risk assessment model, automatic approval of a credit card is realized, and the identification ability of client fraud risk behavior is strengthened.

Description

Technical Field

[0001] This disclosure relates to the field of data processing technology, and in particular to risk prediction methods, apparatus, electronic devices, and storage media. Background Technology

[0002] Currently, in the era of the internet and big data, bank credit cards face severe challenges from fraud risks. As fraud methods become increasingly diversified, professional, and organized, prevention is becoming more difficult, causing extremely serious losses to financial institutions. Mainstream fraud detection methods primarily rely on rules extracted from business experience, which depend on the expertise of business specialists. Based on actual business operations, layers of rules are set up in the approval process to screen fraudulent customers at each stage. By extracting relevant characteristic variables indicating high-risk fraud, corresponding rules are formulated to intercept fraudulent customers.

[0003] The current method of intercepting fraud risks from newly issued cards by fraud rules and manual verification is facing challenges. As the customer base grows, the proportion of suspected fraudulent customers also increases, which hinders the current automated approval process and affects the timeliness of card issuance. Summary of the Invention

[0004] This disclosure provides a method, apparatus, electronic device, and storage medium for risk prediction.

[0005] According to one aspect of this disclosure, a risk prediction method is provided, comprising:

[0006] Obtain the application information and telephone verification information of the target user;

[0007] The first risk assessment result is obtained based on the application information and the electrical and nuclear information using the first risk assessment model;

[0008] A second risk assessment result is obtained based on the application information and the electrical and nuclear information using a second risk assessment model.

[0009] The third risk assessment result is calculated based on the first risk assessment result and the second risk assessment result.

[0010] According to another aspect of this disclosure, a risk prediction device is provided, comprising:

[0011] The acquisition module is configured to acquire the target user's application information and telephone verification information;

[0012] The first assessment module is configured to use the first risk assessment model to obtain a first risk assessment result based on the application information and the electrical and nuclear information.

[0013] The second assessment module is configured to use the second risk assessment model to obtain a second risk assessment result based on the application information and the nuclear power information.

[0014] The comprehensive assessment module is configured to calculate a third risk assessment result based on the first risk assessment result and the second risk assessment result.

[0015] This disclosure also provides an electronic device, including:

[0016] At least one processor; and

[0017] A memory communicatively connected to the at least one processor; wherein,

[0018] The memory stores instructions that can be executed by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform the method described in any of the above technical solutions.

[0019] This disclosure also provides a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to cause the computer to perform the method according to any one of the above embodiments.

[0020] This disclosure also provides a computer program product, including a computer program that, when executed by a processor, implements the method according to any one of the above embodiments.

[0021] This disclosure provides a risk prediction method, apparatus, electronic device, and storage medium. By constructing a first risk assessment model and a second risk assessment model, different business objectives are identified and comprehensively evaluated to enhance the ability to identify customer fraud risk behavior. This is applied to the new customer approval process to improve the ability to identify fraud risk customers during the approval process, optimize the flow of manual telephone verification, and increase the automation rate of telephone verification. The model is developed using big data algorithms to realize the use of quantitative tools to assess the risk of abnormal customer fraud behavior.

[0022] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0023] The accompanying drawings are provided to better understand this solution and do not constitute a limitation of this disclosure. Wherein:

[0024] Figure 1 This is a schematic diagram illustrating the steps of the risk prediction method in an embodiment of this disclosure;

[0025] Figure 2 This is a flowchart illustrating the steps of the model training method in the risk prediction method of this disclosure embodiment;

[0026] Figure 3 This is a schematic block diagram of the risk prediction device in the embodiments of this disclosure;

[0027] Figure 4 This is a schematic diagram of the first risk assessment module in this embodiment of the disclosure;

[0028] Figure 5 This is a schematic diagram of the second risk assessment module in an embodiment of this disclosure;

[0029] Figure 6 This is a block diagram illustrating the principle of the model training module in the risk prediction device according to an embodiment of this disclosure. Detailed Implementation

[0030] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0031] This disclosure provides a risk prediction method, such as Figure 1 As shown, it includes:

[0032] Step S101: Obtain the target user's application information and verification information;

[0033] Step S102: Obtain the first risk assessment result based on the application information and the electrical and nuclear information using the first risk assessment model;

[0034] Step S103: The second risk assessment result is obtained based on the application information and the electrical and nuclear information using the second risk assessment model. It should be noted that the execution order of steps S103 and S102 is not limited. Step S102 can be executed first or step S103 can be executed first, or both can be performed simultaneously.

[0035] Step S104: Calculate the third risk assessment result based on the first and second risk assessment results.

[0036] Specifically, the risk prediction method disclosed herein can be applied to the credit card approval process. In this embodiment, the application information can include information from the application form filled out by the credit card customer, the applicant's device information, credit information from the People's Bank of China, third-party credit information, educational background information, business travel information, retail information within the bank, business registration information, GPS location information, application behavior information, relationship network information, facial recognition information, etc., to comprehensively understand the customer's situation through multiple aspects of information. Telephone verification information refers to customer information obtained during the approval process by making outbound calls to target customers and asking them questions based on preset criteria.

[0037] The first and second risk assessment models target different business objectives. The first risk assessment model is mainly used to identify the abnormality rate of telephone verification, that is, whether the target customer's information is false during the telephone verification process. Abnormal situations in the telephone verification process can generally include: "unscrupulous intermediaries", "fake applications", "fake companies", "fake company information", "fake employment", and "comprehensive abnormality" customers. The second risk assessment model is mainly used to identify the fraud risk rate, mainly to predict the probability of a customer defaulting after applying for a credit card, such as overdue payments after the card is issued.

[0038] Furthermore, the two models will generate a set of assessment results based on their respective identification targets. For example, the probability of a customer's electronic verification being abnormal is 20%, and the probability of fraud is 3%. Based on these two probabilities, a comprehensive risk assessment is conducted to obtain the final risk assessment result. The models can automatically complete credit card approvals, increasing the proportion of automated approvals and improving credit card approval efficiency. On the other hand, simultaneously using two models to assess different business objectives can strengthen the ability to identify customers with fraud risk, minimizing or avoiding bank losses as much as possible.

[0039] As an optional implementation, step S102 specifically includes: the first risk assessment model extracting a first type of feature vector based on the application information and the nuclear power information; and the first risk assessment model obtaining a first risk assessment result based on the first type of feature vector. Step S103 specifically includes: the second risk assessment model extracting a second type of feature vector based on the application information and the nuclear power information; and the second risk assessment model obtaining a second risk assessment result based on the second type of feature vector. Since the first risk assessment model and the second risk assessment model have different identification targets and focus on different feature variables, it is necessary for the first risk assessment model and the second risk assessment model to extract their respective feature vectors for risk analysis based on the nuclear power information and the application information, thereby ensuring the accuracy of the risk assessment.

[0040] As an optional implementation, the first type of feature vector extracted using the first risk assessment model includes at least one of the following: basic information of the application form; credit information from the People's Bank of China; educational background information; third-party credit information; retail information within the bank; application processing behavior information; relationship network information; and applicant's device information. The second type of feature vector extracted using the second risk assessment model includes at least one of the following: basic information of the application form; credit information from the People's Bank of China; educational background information; third-party credit information; retail information within the bank; and application processing behavior information.

[0041] Specifically, in this embodiment, since the identification targets of the first risk assessment model and the second risk assessment model are different, the feature vectors extracted by the two models may be different, and the weights of each feature vector will also be different. For example, the feature vectors extracted by the first risk assessment model may include basic information of the application form, credit information from the People's Bank of China, educational background information, third-party credit information, retail information within the bank, application processing behavior information, relationship network information, and applicant's device information. The proportions of each feature vector in the assessment process are 25.33%, 26.67%, 9.33%, 4.00%, 2.67%, 24.00%, 5.33%, and 2.67%, respectively. The feature vectors extracted by the second risk assessment model may include basic information of the application form, credit information from the People's Bank of China, educational background information, third-party credit information, retail information within the bank, and application processing behavior information. The proportions of each feature vector may be 10.98%, 53.66%, 3.66%, 26.83%, 1.22%, and 3.66%. In the above scheme, the feature variables that the model focuses on will be different depending on the identification target, and the degree of attention paid to each feature variable will also be different.

[0042] As an optional implementation method, the first risk assessment result is calculated based on the first type of feature vector using the first risk assessment model, including:

[0043] The first risk assessment model sorts multiple target users by level based on the first type of feature vectors corresponding to multiple target users, and outputs the probability of electrical core abnormality of target users in each stage as the first risk assessment result.

[0044] The second risk assessment results, calculated using the second type of feature vectors based on the second risk assessment model, include:

[0045] The second risk assessment model ranks multiple target users by level based on the second type of feature vectors corresponding to multiple target users, and outputs the fraud risk probability of the target users in each stage as the second risk assessment result.

[0046] For example, Table 1 shows the risk ranking performance of the first risk assessment model in the new customer group, i.e., the first risk assessment result. In this embodiment, the tiered ranking means that multiple new credit card applicants are divided into 10 stages during the assessment process. The top 10% of customers are high-risk customers, and their corresponding abnormal call verification probability is 25.71%. For example, if there are 100 new credit card applicants, divided into 10 groups, the 10 new customers in the top 10% have the highest abnormal call verification probability. The abnormal call verification probability of the 10 new customers in the 10% to 20% group drops to 14.78%, and the abnormal call verification probability of the 20% to 30% group further drops to 9.44%. The abnormal call verification probability in Table 1 decreases from top to bottom, and the risk probability decreases in a step-like manner. The abnormal call verification probability of customers in the 90% to 100% stage is the lowest, at only 0.12%.

[0047]

[0048] Table 1. Risk Ranking of New Customer Groups by the First Risk Assessment Model

[0049] Table 2 shows the risk ranking performance of the second risk assessment model in the new customer group, which is the result of the first risk assessment. Similarly, in this embodiment, the second risk assessment model will also be used to rank the target customers by risk. First, multiple new customers applying for credit cards are divided into 10 stages. The customers in the top 10% are high-risk customers, and their corresponding fraud risk probability is 1.97%. The fraud risk probability in Table 2 decreases from top to bottom. Among them, the fraud risk probability of customers in the 90% to 100% stage is the lowest, about 0.03%.

[0050]

[0051] Table 2. Risk Ranking of New Customer Groups by the Second Risk Assessment Model

[0052] Through the above technical solution, the risk assessment range of the target customer is obtained through two risk assessment models, yielding the corresponding probability of abnormal phone verification and fraud risk. A comprehensive assessment can be further conducted based on the results of these two aspects to obtain a comprehensive score, which serves as the final third risk assessment result. A first threshold can be set for the first risk assessment result, and a second threshold can be set for the second risk assessment result. For example, the first threshold could be set to 10%, and the second threshold to 1%. When the probability of abnormal phone verification is greater than or equal to 10%, the system determines that the customer has an abnormal phone verification, meaning they have not passed the assessment of the first risk assessment model and are classified as a high-risk customer, thus their credit card application will not be approved. And / or when the probability of fraud risk is greater than or equal to 1%, the system determines that the customer has a fraud risk, meaning they have not passed the assessment of the second risk assessment model and are classified as a high-risk customer, thus their credit card application will not be approved. A third threshold can also be set for the third risk assessment result. If the risk score represented by the third risk assessment result does not reach the third threshold, the customer's credit card application will not be approved. This method enables automated credit card approval processes, improving the efficiency of credit card approval; and by conducting comprehensive evaluation from two aspects, it can enhance the ability to identify customer fraud risk behaviors and minimize bank losses.

[0053] As an optional implementation, step S104, which calculates the third risk assessment result based on the first and second risk assessment results, includes:

[0054] Based on the probability of nuclear anomalies, the first good-to-bad ratio corresponding to the first risk assessment result is determined, and the first good-to-bad ratio is converted into the corresponding first score.

[0055] The second good-bad ratio corresponding to the second risk assessment result is determined based on the fraud risk probability, and the second good-bad ratio is converted into the corresponding second score.

[0056] The total score, derived from the first and second scores, serves as the third risk assessment result.

[0057] Assuming that the risk characteristics of historical customers remain largely consistent with future applications, based on big data principles, we analyze various information such as attribute information, credit information, and device comparison information. By analyzing the relationship between these attributes of historical customers and the good and bad event rates, we predict the probability of good or bad new customers and convert the probability into a score to facilitate business applications.

[0058] Specifically, this embodiment uses a nonlinear regression model to fit the linear relationship between the total score and the logarithm of the good-to-bad ratio. The independent variable of the regression model is the total score, and the dependent variable is the logarithm of the good-to-bad ratio. The probabilities predicted by the model are converted into scores using the following formula:

[0059]

[0060] Where G represents the good customer sample; B represents the bad customer sample; S represents the total sample; Pr{G|S} represents the proportion of the good customer sample; Pr{B|S} represents the proportion of the bad customer sample; and Odds represents the good-bad ratio.

[0061] Furthermore, the first and second good-bad ratios are converted into corresponding first and second scores using the following formulas:

[0062] S = score + PDO * (ln(Odds) - ln(baseline Odds)) / ln(2)

[0063] Here, score represents the baseline score; Odds represents the good-to-bad ratio; and PDO represents the additional score required to double the good-to-bad ratio.

[0064] Replace it with the mapping function learned by the XGBoost model. The XGBoost model uses CART (Classification And Regression Trees) regression trees as weak learners, where y represents the error and ft represents each regression tree.

[0065] As an optional implementation, this disclosure also includes a method for training a risk assessment model, which further includes pre-training a first risk assessment model and a second risk assessment model before obtaining the application information and power information of the target user for risk assessment:

[0066] Step S201: Obtain reference feature variables based on historical application information and historical nuclear power information;

[0067] Step S202: Preprocess the reference feature variables;

[0068] Step S203: Filter the reference feature variables;

[0069] Step S204: Obtain modeling samples and test samples as training samples, and extract training feature variables from the training samples based on reference feature variables;

[0070] Step S205: Train the first risk assessment model and the second risk assessment model based on the training feature variables extracted from the training samples.

[0071] Specifically, the first and second risk assessment models are applied to the approval process in new credit card issuance, therefore, the models are applicable to new credit card applicants. Since the risk profiles of special customer groups not applicable to the models differ significantly from those of the normal customer group, it is necessary to exclude data from historical customer data during the modeling process. Combining risk management business policies and special customer group rules for incoming applications, the exclusion rules for the first risk assessment model are as follows: applications for Blue Cards, supplementary cards, existing cards, test cards, and applications without electronic verification or verification are not included in the modeling scope; the exclusion rules for the second risk assessment model are as follows: applications for Blue Cards, supplementary cards, existing cards, test cards, and applications from rejected customers are not included in the modeling scope.

[0072] Since the first and second risk assessment models identify different targets, it is necessary to define their respective identification targets. Table 3 lists the target customers to be identified by the first risk assessment model, and Table 4 lists the target customers to be identified by the second risk assessment model.

[0073]

[0074] Table 3. Identification Targets of the First Risk Assessment Model

[0075]

[0076] Table 4. Identification Targets of the Second Risk Assessment Model

[0077] Specifically, in step S201, the first step is to obtain the feature variables needed for modeling. This involves in-depth mining of customer information, including basic application information, applicant device information, People's Bank of China credit information, third-party credit information, educational background information, business travel information, retail information within the industry, business registration information, GPS location information, application behavior, relationship network information, facial recognition information, etc., striving to fully utilize existing data to find and quantify the risk characteristics of credit card business from multiple perspectives, thereby improving the predictive ability of the final model.

[0078] During the process of acquiring feature variables, analyzing the date field can reveal the time period of the downloaded data, thus providing a preliminary assessment of whether the downloaded data meets project requirements and aligns with business understanding. Checking data integrity is crucial. Primary key (i.e., the primary key, a candidate key selected to uniquely identify rows in a table) uniqueness checks are a vital part of data quality control and an indispensable step in subsequent data association analysis and data mart construction. Therefore, it is necessary to confirm the unique keys of important tables.

[0079] In step S202, the acquired feature variables need to be preprocessed, including cleaning and derivation. The reasons for missing values ​​in the feature variables are analyzed for their business implications, and specific values ​​are assigned to represent these meanings, as shown in Table 5.

[0080] Value definition -9999999 This special value indicates that such information is unavailable. -9999989 No human rights report

[0081] Table 5. Handling of missing values ​​for feature variables

[0082] During model development, one-hot encoding was used to transform character-type indicator variables. After one-hot encoding, several variable value features are transformed from one variable to n feature variables. For example, the "gender" indicator, using unique OneHot encoding, can generate two columns: "male" and "female". Ultimately, one-hot encoding can transform discrete features into continuous features. Descriptive statistical analysis of candidate variables can reflect the general distribution of the variables as a whole. After handling missing values ​​for all candidate variables according to their business meaning, descriptive statistical analysis is required.

[0083] After preprocessing the feature variables, step S203 involves screening the feature variables and performing value (IV) analysis, correlation analysis, and cross-time point PSI (Population Stability Index) analysis. IV analysis identifies feature variables with strong predictive power for the model's objective; correlation analysis filters out variables with high correlations; and PSI analysis removes variables with high PSI values ​​to avoid affecting model stability.

[0084] In step S204, training samples are obtained from historical credit card customer data. During the training of the risk assessment model, both modeling samples and test samples need to be obtained simultaneously. The modeling samples are used for initial modeling, and after the model is established, test samples are needed to verify the model's accuracy and stability. In this embodiment, the timing of the modeling samples and test samples in the development of the first risk assessment model can be determined based on the distribution of the sample data and combined with business experience, as shown in Table 6.

[0085] Sample classification Observation period Performance period Modeling Samples 201908-201910 201911-202001 Validation Sample 201911-202001 202002-202004

[0086] Table 6. Modeling and Testing Sample Time Points in the Development of the First Risk Assessment Model

[0087] The performance period of the first risk assessment model is three months after the observation point. For applications submitted in August 2019, the performance window is from August 2019 to November 2019. For example, we selected customer application data from August to October 2019, and the performance period is the performance of these customers in the following three months, i.e., whether there were any bad events such as fraud or default during November 2019 to January 2020. In addition, since the obtained sample data may be imbalanced, that is, the ratio of good and bad samples may be severely skewed to one side, generally bad samples will be significantly less than good samples, it is necessary to address the sample imbalance problem in the modeling. In this embodiment, validation samples from different time periods were obtained to ensure the accuracy of the model.

[0088] Similarly, the second risk assessment model also needs to determine the modeling sample time point and the test sample time point during development. The performance period is 4 months after the observation time point. For applications submitted in January 2020, the performance window is from January 2020 to April 2020, as shown in Table 7.

[0089] Sample classification Observation period Performance period Modeling Samples 202003-202005 202006-202009 Validation Sample 201911-202002 202003-202006

[0090] Table 7. Modeling Sample Time Points and Testing Sample Time Points in the Development of the Second Risk Assessment Model

[0091] After cleaning and processing the feature variables and preparing the sample data for both models, machine learning modeling was performed. Based on the training samples obtained in step S204, a machine learning algorithm (XgBoost algorithm) was used for modeling. Assuming that the risk characteristics of historical customers remain basically consistent with future applications, and based on the law of large numbers, various information such as attribute information, credit information, and device comparison information were analyzed. Based on the relationship between the above attributes of historical customers and the good and bad event rates (prior probabilities), the probability of good or bad customers in the future was predicted, and the probability was converted into a score to facilitate business applications.

[0092] This disclosure also provides a risk prediction device, such as Figure 3 As shown, it includes:

[0093] The acquisition module 301 is configured to acquire the application information and telephone verification information of the target user.

[0094] The first assessment module 302 is configured to use the first risk assessment model to obtain the first risk assessment result based on the application information and the electrical and nuclear information.

[0095] The second assessment module 303 is configured to use the second risk assessment model to obtain the second risk assessment result based on the application information and the electrical and nuclear information.

[0096] The comprehensive assessment module 304 is configured to calculate the third risk assessment result based on the first risk assessment result and the second risk assessment result.

[0097] Specifically, the risk prediction device disclosed herein can be applied to the credit card approval process. In this embodiment, the application information can include information from the application form filled out by the credit card customer, applicant device information, People's Bank of China credit information, third-party credit information, educational background information, business travel information, retail information within the bank, business registration information, GPS location information, application behavior information, relationship network information, facial recognition information, etc., to comprehensively understand the customer's situation through multiple aspects of information. Telephone verification information refers to customer information obtained by making outbound calls to target customers during the approval process and asking them questions based on preset questions.

[0098] The first and second risk assessment models target different business objectives. The first risk assessment model is mainly used to identify the abnormality rate of telephone verification, that is, whether the target customer's information is false during the telephone verification process. Abnormal situations in the telephone verification process can generally include: "unscrupulous intermediaries", "fake applications", "fake companies", "fake company information", "fake employment", and "comprehensive abnormality" customers. The second risk assessment model is mainly used to identify the fraud risk rate, mainly to predict the probability of a customer defaulting after applying for a credit card, such as overdue payments after the card is issued.

[0099] Furthermore, the two models will generate a set of assessment results based on their respective identification targets. For example, the probability of a customer's electronic verification being abnormal is 20%, and the probability of fraud is 3%. Based on these two probabilities, a comprehensive risk assessment is conducted to obtain the final risk assessment result. The models can automatically complete credit card approvals, increasing the proportion of automated approvals and improving credit card approval efficiency. On the other hand, simultaneously using two models to assess different business objectives can strengthen the ability to identify customers with fraud risk, minimizing or avoiding bank losses as much as possible.

[0100] As an optional implementation method, such as Figure 4 As shown, the first evaluation module 302 includes:

[0101] The first feature extraction unit 3021 is configured to extract a first type of feature vector based on application information and nuclear information using a first risk assessment model.

[0102] The first computing unit 3022 is configured to obtain a first risk assessment result based on a first type of feature vector;

[0103] like Figure 5 As shown, the second evaluation module 303 includes:

[0104] The second feature extraction unit 3031 is configured to extract a second type of feature vector based on application information and nuclear information using a second risk assessment model.

[0105] The second calculation unit 3032 is configured to obtain the second risk assessment result based on the second type of feature vector.

[0106] Specifically, since the first risk assessment model and the second risk assessment model have different identification targets, they also focus on different feature variables. Therefore, the first risk assessment model and the second risk assessment model need to extract their respective feature vectors for risk analysis based on the nuclear information and application information, so as to ensure the accuracy of risk assessment.

[0107] As an optional implementation method, such as Figure 6 As shown, the risk prediction device also includes a model training module 305, configured to train a first risk assessment model and a second risk assessment model. The model training module includes:

[0108] The first acquisition unit 3051 is configured to acquire reference feature variables based on historical application information and historical nuclear power information;

[0109] Preprocessing unit 3052 is configured to preprocess the reference feature variables;

[0110] The filtering unit 3053 is configured to filter reference feature variables;

[0111] The second acquisition unit 3054 is configured to acquire modeling samples and test samples as training samples, and extract training feature variables from the training samples based on reference feature variables.

[0112] Training unit 3055 is configured to train the first risk assessment model and the second risk assessment model based on the training feature variables extracted from the training samples.

[0113] Specifically, the model training module in this embodiment is the same as the model training method in the above embodiments, so it will not be described again below.

[0114] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0115] Specifically, electronic devices are intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. Electronic devices can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0116] The device includes a computing unit that can perform various appropriate actions and processes based on a computer program stored in read-only memory (ROM) or loaded from a storage unit into random access memory (RAM). The RAM can also store various programs and data required for device operation. The computing unit, ROM, and RAM are interconnected via a bus. Input / output (I / O) interfaces are also connected to the bus.

[0117] Multiple components in the device are connected to the I / O interface, including: input units such as keyboards and mice; output units such as various types of displays and speakers; storage units such as disks and optical discs; and communication units such as network interface cards (NICs), modems, and wireless transceivers. The communication unit allows the device to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0118] The computing unit can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of computing units include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit performs the various methods and processes described above, such as the risk prediction method in the above embodiments. For example, in some embodiments, the risk prediction method may be implemented as a computer software program tangibly contained in a machine-readable medium, such as a storage unit. In some embodiments, part or all of the computer program may be loaded and / or installed on the device via ROM and / or a communication unit. When the computer program is loaded into RAM and executed by the computing unit, one or more steps of the risk prediction method described above may be performed. Alternatively, in other embodiments, the computing unit may be configured to perform the risk prediction method by any other suitable means (e.g., by means of firmware).

[0119] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0120] The program code used to implement the risk prediction method of this disclosure can be written in any combination of one or more programming languages. This program code can be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing device, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code can be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0121] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0122] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0123] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with embodiments of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0124] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, servers in distributed systems, or servers incorporating blockchain technology.

[0125] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0126] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A risk prediction method, characterized in that, include: During the credit card approval process, obtain the target user's application information and electronic verification information; The first risk assessment model is used to extract a first type of feature vector based on the application information and the power information. Based on the first type of feature vector corresponding to the multiple target users, the multiple target users are sorted in order of order, and the power abnormality probability of the target users in each stage is output as the first risk assessment result. The second risk assessment model is used to extract a second type of feature vector based on the application information and the electronic information. Based on the second type of feature vector corresponding to the multiple target users, the multiple target users are sorted in order of priority. The fraud risk probability of the target users in each stage is output as the second risk assessment result. The first type of feature vector includes relationship network information and applicant device information. The second type of feature vector does not include the above information. Based on the probability of nuclear anomaly, a first good-to-bad ratio corresponding to the first risk assessment result is determined, and the first good-to-bad ratio is converted into a corresponding first score; based on the probability of fraud risk, a second good-to-bad ratio corresponding to the second risk assessment result is determined, and the second good-to-bad ratio is converted into a corresponding second score; based on the first score and the second score, a total score is obtained as the third risk assessment result; The first good-to-bad ratio and the second good-to-bad ratio are calculated using the following formulas: Where G represents good customer samples; B represents bad customer samples; S represents all samples; Pr{G|S} represents the proportion of good customer samples; Pr{B|S} represents the proportion of bad customer samples; Odds represents the good-to-bad ratio; the first good-to-bad ratio and the second good-to-bad ratio are converted into the corresponding first and second scores through the mapping function learned by the XGBoost model; the XGBoost model uses a classification regression tree as a weak learner.

2. The method according to claim 1, characterized in that, The first type of feature vector includes at least one of the following: basic information of the application form; credit information from the People's Bank of China; educational background information; third-party credit information; retail information within the bank; application behavior information; relationship network information; and applicant's device information.

3. The method according to claim 1, characterized in that, The second type of feature vector includes at least one of the following: basic information of the application form; credit information from the People's Bank of China; educational background information; third-party credit information; retail information within the bank; and application processing information.

4. The method according to claim 1, characterized in that, The first good-bad ratio and the second good-bad ratio are converted into the corresponding first score and second score using the following formulas: S = score + PDO (ln(Odds)-ln(baseline Odds)) / ln(2) Wherein, score represents the baseline score; Odds represents the good-to-bad ratio; and PDO represents the additional score required to double the good-to-bad ratio.

5. The method according to claim 1, characterized in that, Before obtaining the target user's application information and telephone verification information, the process also includes pre-training the first risk assessment model and the second risk assessment model: Reference feature variables are obtained based on historical application information and historical nuclear power information; The reference feature variables are preprocessed; The reference feature variables are then filtered. Obtain modeling samples and test samples as training samples, and extract training feature variables from the training samples based on the reference feature variables; The first risk assessment model and the second risk assessment model are obtained by training the training feature variables extracted from the training samples.

6. A risk prediction device, characterized in that, include: The acquisition module is configured to acquire the target user's application information and electronic verification information during the credit card approval process; The first assessment module is configured to use a first risk assessment model to extract a first type of feature vector based on the application information and the electrical core information, and to sort the multiple target users in order based on the first type of feature vector corresponding to the multiple target users, and output the electrical core abnormality probability of the target users in each stage as the first risk assessment result. The second assessment module is configured to use a second risk assessment model to extract a second type of feature vector based on the application information and the electronic information, and to sort the multiple target users in order of priority based on the second type of feature vector corresponding to the multiple target users, and output the fraud risk probability of the target users in each stage as the second risk assessment result; wherein, the first type of feature vector includes relationship network information and applicant device information; the second type of feature vector does not include the above information; The comprehensive assessment module is configured to determine a first good-to-bad ratio corresponding to the first risk assessment result based on the probability of electrical core anomaly, and convert the first good-to-bad ratio into a corresponding first score; determine a second good-to-bad ratio corresponding to the second risk assessment result based on the probability of fraud risk, and convert the second good-to-bad ratio into a corresponding second score; and obtain a total score based on the first score and the second score as a third risk assessment result. The first good-to-bad ratio and the second good-to-bad ratio are calculated using the following formulas: Where G represents good customer samples; B represents bad customer samples; S represents all samples; Pr{G|S} represents the proportion of good customer samples; Pr{B|S} represents the proportion of bad customer samples; Odds represents the good-to-bad ratio; the first good-to-bad ratio and the second good-to-bad ratio are converted into the corresponding first and second scores through the mapping function learned by the XGBoost model; the XGBoost model uses a classification regression tree as a weak learner.

7. The risk prediction device according to claim 6, characterized in that, It also includes a model training module configured to train the first risk assessment model and the second risk assessment model, the model training module comprising: The first acquisition unit is configured to acquire reference feature variables based on historical application information and historical nuclear power information; The preprocessing unit is configured to preprocess the reference feature variables; The filtering unit is configured to filter the reference feature variables; The second acquisition unit is configured to acquire modeling samples and test samples as training samples, and extract training feature variables from the training samples based on the reference feature variables. The training unit is configured to train the first risk assessment model and the second risk assessment model based on the training feature variables extracted from the training samples.

8. An electronic device, comprising: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-5.

9. A non-transitory computer-readable storage medium storing computer instructions, wherein, The computer instructions are used to cause the computer to perform the method according to any one of claims 1-5.

10. A computer program product comprising a computer program that, when executed by a processor, implements the method according to any one of claims 1-5.

Images