Business data feature completion, model training and risk prediction method and device

Abstract

The invention relates to the technical field of data processing, and provides a business data feature completion method and device, a model training method and device and a risk prediction method and device, and the filling method comprises the steps: obtaining a first variable of business features in a business data set and a second variable which is not randomly missing; training to obtain a business prediction basic model based on the sample of the first variable, and generating a business prediction possibility sequence according to the business prediction basic model; determining a deletion type of a non-random deletion second variable, and determining a sample selection range of the second variable according to the deletion type; and feature completion processing is performed on a target sample according to the service prediction possibility sequence, so that an obtained second variable new sample conforms to the possibility distribution of the second variable, and the target sample is a parameter value of the second variable of the sample located in the sample selection range. According to the embodiment of the invention, the accuracy of business data feature completion can be improved, the prediction accuracy of a business risk prediction model is improved, and the accuracy of business risk prediction is improved.

Description

Technical Field

[0001] This specification relates to the technical field of data processing, and particularly to a method and apparatus for completing business data features, training a model, and predicting risks. Background Art

[0002] With the development and application of artificial intelligence technology, there have currently emerged technical solutions for predicting business risks using artificial intelligence technology. For example, taking security risks in the fintech field (such as credit risk and account anomaly risk) as an example, a pre-trained risk prediction model can be used to predict the anomaly risk (such as default risk) of users. However, when training a model, sometimes there is a problem of missing sample features of variables (such as missing variables), which can easily affect the prediction accuracy of using the risk prediction model. Summary of the Invention

[0003] The purpose of the embodiments of this specification is to provide a method and apparatus for completing business data features, training a model, and predicting risks, so as to improve the accuracy of completing business data features, improve the prediction accuracy of the business risk prediction model, and improve the accuracy of business risk prediction.

[0004] To achieve the above purpose, on the one hand, the embodiments of this specification provide a method for completing business data features, including:

[0005] Obtain the first variable of the business feature and the second variable of non-random missing in the business dataset;

[0006] Based on the samples of the first variable, train a business prediction basic model, and generate a business prediction possibility sequence according to the business prediction basic model;

[0007] Determine the missing type of the non-random missing second variable, and determine the sample selection range of the second variable according to the missing type;

[0008] Perform feature completion processing on the target sample according to the business prediction possibility sequence, so that the new sample of the second variable obtained conforms to the probability distribution of the second variable, and the target sample is the parameter value of the second variable of the sample within the sample selection range.

[0009] In the method for completing business data features of the embodiments of this specification, the performing feature completion processing on the target sample according to the business prediction possibility sequence includes:

[0010] Convert the target sample to obtain a specified type parameter sequence of the second variable;

[0011] Determine the cumulative distribution sequence of the specified type parameter sequence of the second variable;

[0012] Determine the overall distribution of the service prediction possibility sequence according to the cumulative distribution sequence;

[0013] Generate a mapping relationship according to the overall distribution and the specified type parameter sequence of the second variable;

[0014] Input the service prediction possibility sequence into the mapping relationship to obtain a corresponding new sample of the second variable.

[0015] In the method for completing service data feature in the embodiment of this specification, determining the overall distribution of the service prediction possibility sequence according to the cumulative distribution sequence includes:

[0016] Perform a sorting process on the cumulative distribution sequence from small to large;

[0017] Perform cumulative distribution completion processing on both ends of the sorted cumulative distribution sequence;

[0018] Convert the cumulative distribution sequence after cumulative distribution completion into the overall distribution of the service prediction possibility sequence.

[0019] In the method for completing service data feature in the embodiment of this specification, generating a mapping relationship according to the overall distribution and the specified type parameter sequence of the second variable includes:

[0020] Perform a slicing process on the value range of the service prediction possibility sequence according to the overall distribution to obtain multiple first value ranges;

[0021] Generate a corresponding second value range for each of the first value ranges, where the second value range is an integer falling within the target range;

[0022] For each first value range, determine the union of each first value range and the corresponding second value range of each first value range as the mapping relationship.

[0023] In the method for completing service data feature in the embodiment of this specification, generating a service prediction possibility sequence according to the service prediction basic model includes:

[0024] Input a specified part of the samples in the first variable into the service prediction basic model to obtain a corresponding service prediction possibility sequence.

[0025] In the method for completing service data feature in the embodiment of this specification, determining the sample selection range of the second variable according to the missing type includes:

[0026] When the type of the non-randomly missing second variable is the first type, the sample selection range of the second variable is to select all samples corresponding to the second variable from the service dataset; or,

[0027] When the type of the second variable with non-random missingness is the second type, the sample selection range of the second variable is to select samples corresponding to the second variable that conform to the truncation rule from the business dataset;

[0028] Wherein, the first type represents non-random missingness caused by the retrospective time limit, and the second type represents non-random missingness caused by the truncation rule limit.

[0029] On the other hand, an embodiment of this specification also provides a method for training a business risk prediction model, including:

[0030] Obtain the first variable of the business features and the second variable with non-random missingness in the business dataset;

[0031] Based on the samples of the first variable, train a first business risk prediction model, and generate a business risk possibility sequence according to the first business risk prediction model;

[0032] Determine the missing type of the non-random missing second variable, and determine the sample selection range of the second variable according to the missing type;

[0033] Perform feature completion processing on the target samples according to the business risk possibility sequence, so that the new samples of the second variable obtained conform to the probability distribution of the second variable, where the target samples are the parameter values of the second variable of the samples within the sample selection range;

[0034] Train a second business risk prediction model based on the samples of the first variable, the new samples of the second variable, and the samples of the non-random missing second variable.

[0035] In the method for training a business risk prediction model according to the embodiment of this specification, the performing feature completion processing on the target samples according to the business risk possibility sequence includes:

[0036] Perform conversion processing on the target samples to obtain a specified type parameter sequence of the second variable;

[0037] Determine the cumulative distribution sequence of the specified type parameter sequence of the second variable;

[0038] Determine the overall distribution of the business risk possibility sequence according to the cumulative distribution sequence;

[0039] Generate a mapping relationship according to the overall distribution and the specified type parameter sequence of the second variable;

[0040] Input the business risk possibility sequence into the mapping relationship to obtain the corresponding new samples of the second variable.

[0041] In the method for training a business risk prediction model according to an embodiment of this specification, determining the overall distribution of the business risk possibility sequence based on the cumulative distribution sequence includes:

[0042] Performing a sorting process on the cumulative distribution sequence from smallest to largest;

[0043] Performing cumulative distribution completion processing on both ends of the sorted cumulative distribution sequence;

[0044] Converting the cumulative distribution sequence after cumulative distribution completion into the overall distribution of the business risk possibility sequence.

[0045] In the method for training a business risk prediction model according to an embodiment of this specification, generating a mapping relationship based on the overall distribution and the specified type parameter sequence of the second variable includes:

[0046] Performing a slicing process on the value range of the business risk possibility sequence according to the overall distribution to obtain a plurality of first value ranges;

[0047] Generating a corresponding second value range for each of the first value ranges, where the second value range is an integer falling within the target range;

[0048] For each first value range, determining the union of each first value range and the corresponding second value range of each first value range as the mapping relationship.

[0049] In the method for training a business risk prediction model according to an embodiment of this specification, generating a business risk possibility sequence based on the first business risk prediction model includes:

[0050] Inputting some samples in the first variable into the first business risk prediction model to obtain a corresponding business risk possibility sequence.

[0051] In the method for training a business risk prediction model according to an embodiment of this specification, determining the sample selection range of the second variable according to the missing type includes:

[0052] When the type of the non-randomly missing second variable is the first type, the sample selection range of the second variable is to select all samples corresponding to the second variable from the business dataset; or,

[0053] When the type of the non-randomly missing second variable is the second type, the sample selection range of the second variable is to select samples corresponding to the second variable from the business dataset that conform to the truncation rule;

[0054] Among them, the first type represents non-random missingness caused by the backtracking time limit, and the second type represents non-random missingness caused by the truncation rule limit.

[0055] On the other hand, an embodiment of this specification also provides a business risk prediction method, including:

[0056] Obtain the business data of the target object, where the business data includes the first variable data and the second variable data of the business characteristics of the target object;

[0057] Input the business data into a pre-trained business risk prediction model to obtain the business risk possibility of the target object; the business risk prediction model includes the second business risk prediction model obtained by the above business risk prediction model training method.

[0058] On the other hand, an embodiment of this specification also provides a business data feature completion device, including:

[0059] A first acquisition module, configured to acquire the first variable of the business characteristics and the second variable with non-random missingness in the business dataset;

[0060] A first generation module, configured to train a business prediction basic model based on the samples of the first variable, and generate a business prediction possibility sequence according to the business prediction basic model;

[0061] A first determination module, configured to determine the missing type of the second variable with non-random missingness, and determine the sample selection range of the second variable according to the missing type;

[0062] A first feature completion module, configured to perform feature completion processing on the target sample according to the business prediction possibility sequence, so that the obtained new sample of the second variable conforms to the possibility distribution of the second variable, where the target sample is the parameter value of the second variable of the sample within the sample selection range.

[0063] On the other hand, an embodiment of this specification also provides a business risk prediction model training device, including:

[0064] A second acquisition module, configured to acquire the first variable of the business characteristics and the second variable with non-random missingness in the business dataset;

[0065] A second generation module, configured to train a first business risk prediction model based on the samples of the first variable, and generate a business risk possibility sequence according to the first business risk prediction model;

[0066] A second determination module, configured to determine the missing type of the second variable with non-random missingness, and determine the sample selection range of the second variable according to the missing type;

[0067] A second feature completion module, configured to perform feature completion processing on a target sample according to a sequence of business risk possibilities, so that the obtained new second-variable sample conforms to the probability distribution of the second variable, where the target sample is the parameter value of the second variable of the sample within the sample selection range;

[0068] A training module, configured to train a second business risk prediction model according to the sample of the first variable, the new second-variable sample, and the sample of the second variable with non-random missing values.

[0069] On the other hand, an embodiment of this specification further provides a business risk prediction device, including:

[0070] A third acquisition module, configured to acquire business data of a target object, where the business data includes first-variable data and second-variable data of business features of the target object;

[0071] A risk prediction module, configured to input the business data into a pre-trained business risk prediction model to obtain the business risk possibility of the target object; the business risk prediction model is the second business risk prediction model obtained according to the above business risk prediction model training method.

[0072] On the other hand, an embodiment of this specification further provides a computer device, including a memory, a processor, and a computer program stored on the memory, where when the computer program is run by the processor, it executes the instructions of the above method.

[0073] On the other hand, an embodiment of this specification further provides a computer storage medium, on which a computer program is stored, where when the computer program is run by a processor of a computer device, it executes the instructions of the above method.

[0074] On the other hand, an embodiment of this specification further provides a computer program product, where the computer program product includes a computer program, and when the computer program is run by a processor of a computer device, it executes the instructions of the above method.

[0075] As can be seen from the technical solutions provided in the embodiments of this specification above, in the embodiments of this specification, since the variables used when training the business prediction basic model are all first variables (normal variables) without missing values, and the probability distributions of the samples of these first variables conform to the normal distribution, the probability distribution of the business prediction probability sequence generated based on the samples of the first variables and the business prediction basic model also conforms to the normal distribution. Therefore, by mapping the business prediction probability sequence to the probability distribution mode of the second variable (key variable) without random missing values to generate new samples of the second variable (that is, performing feature completion processing on the target sample according to the business prediction probability sequence), it can be ensured that the probability distribution of the new samples of the second variable obtained by the feature completion processing also conforms to the normal distribution (that is, is consistent with the normal probability distribution of the real samples of the second variable). Therefore, the new samples of the second variable obtained by completion complement the sample features of the second variable caused by non-random missing values, thereby improving the accuracy of business data feature completion; further, when the second variable without random missing values is completed according to the business data feature completion solution of this application and the business risk prediction model is trained on this basis, it is beneficial to improve the prediction accuracy of the trained business risk prediction model, and ultimately beneficial to improving the accuracy of business risk prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] In order to more clearly illustrate the technical solutions in the embodiments of this specification or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments recorded in this specification. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings. In the drawings:

[0077] Figure 1 The flowchart of the business data feature completion method in some embodiments of this specification is shown;

[0078] Figure 2 Shows Figure 1 The flowchart of obtaining the normal variables and non-randomly missing key variables in the method shown;

[0079] Figure 3 Shows Figure 1 The flowchart of the feature completion processing in the method shown;

[0080] Figure 4 Shows Figure 1 The flowchart of determining the percentile of the business prediction probability sequence in the method shown;

[0081] Figure 5 Shows Figure 1Flowchart of generating a mapping relationship based on percentiles and an array of integer values of key variables in the method shown;

[0082] Figure 6 Schematic diagram comparing the probability density distributions of the true and fitted samples of key variables with non - random missing values in an exemplary embodiment of this specification;

[0083] Figure 7 Schematic diagram of the probability density distribution of the full set of key variable X after feature completion processing of key variable X with non - random missing values in an exemplary embodiment of this specification;

[0084] Figure 8a Schematic diagram of the probability density distribution of the full set of key variable X after fitting key variable X with non - random missing values using a traditional fitting function;

[0085] Figure 8b Schematic diagram of the probability density distribution of the full set of key variable X after filling key variable X with non - random missing values using the traditional method of filling missing values with random numbers;

[0086] Figure 8c Schematic diagram of the probability density distribution of the full set of key variable X after filling key variable X with non - random missing values using the traditional method of filling missing values with 0;

[0087] Figure 9 Flowchart of the business risk prediction model training method in some embodiments of this specification;

[0088] Figure 10 Flowchart of the business risk prediction method in some embodiments of this specification;

[0089] Figure 11 Block diagram of the structure of the business data feature completion device in some embodiments of this specification;

[0090] Figure 12 Block diagram of the structure of the business risk prediction model training device in some embodiments of this specification;

[0091] Figure 13 Block diagram of the structure of the business risk prediction device in some embodiments of this specification;

[0092] Figure 14 Block diagram of the structure of the computer device in some embodiments of this specification.

[0093]

Description of Reference Numerals

[0094] 111. First acquisition module;

[0095] 112. First generation module;

[0096] 113. First determination module;

[0097] 114. First feature completion module;

[0098] 121. Second acquisition module;

[0099] 122. Second generation module;

[0100] 123. Second determination module;

[0101] 124. Second feature completion module;

[0102] 125. Training module;

[0103] 131. Third acquisition module;

[0104] 132. Risk prediction module;

[0105] 1402. Computer device;

[0106] 1404. Processor;

[0107] 1406. Memory;

[0108] 1408. Driving mechanism;

[0109] 1410. Input / output interface;

[0110] 1412. Input device;

[0111] 1414. Output device;

[0112] 1416. Presentation device;

[0113] 1418. Graphical user interface;

[0114] 1420. Network interface;

[0115] 1422. Communication link;

[0116] 1424. Communication bus. Detailed implementation manners

[0117] In order to enable those skilled in the art of this technology to better understand the technical solutions in this specification, the following will clearly and completely describe the technical solutions in the embodiments of this specification in conjunction with the accompanying drawings in the embodiments of this specification. Obviously, the described embodiments are only a part of the embodiments of this specification, rather than all the embodiments. Based on the embodiments in this specification, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the scope of protection of this specification.

[0118] It should be noted that in the embodiments of this specification, the user information involved (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) are all information and data that have been authorized and consented to by the user and fully authorized by all parties. That is, in the technical solution of this application, the acquisition, transmission, storage, use, processing, etc. of data all comply with the relevant provisions of national laws and regulations.

[0119] Generally, the variable missing of business data can be divided into two types: Missing At Random (MAR) and Missing Not At Random (MNAR). For random missing, traditional imputation methods (such as mean imputation, median imputation, mode imputation) can better complete the missing features of the sample. However, for non-random missing, since non-random missing will introduce systematic bias, making the variable distribution deviate from its true state, and traditional imputation methods cannot eliminate systematic bias. Especially for key variables, once non-random missing problems occur in key variables, it will significantly affect the prediction accuracy of the credit risk prediction model, and further affect the accuracy of credit risk prediction.

[0120] In view of this, the embodiments of this specification provide a method and device for business data feature completion, model training, and risk prediction to improve the accuracy of business data feature completion, improve the prediction accuracy of the business risk prediction model, and improve the accuracy of business risk prediction.

[0121] Figure 1 The flowchart of the business data feature completion method in some embodiments of this specification is shown. This business data feature completion method can be applied to the computer device side. Refer to Figure 1 As shown, in some embodiments of this specification, the business data feature completion method may include the following steps 101 to 104.

[0122] Step 101: Obtain the first variable of the business feature and the second variable of non-random missing in the business data set.

[0123] Step 102: Based on the samples of the first variable, train a business prediction basic model, and generate a business prediction possibility sequence according to the business prediction basic model.

[0124] Step 103: Determine the missing type of the second variable of non-random missing, and determine the sample selection range of the second variable according to the missing type.

[0125] Step 104. Perform feature completion processing on the target sample according to the business prediction possibility sequence, so that the obtained new sample of the second variable conforms to the possibility distribution of the second variable. The target sample is the parameter value of the second variable of the sample within the sample selection range.

[0126] In the embodiments of this specification, the first variable is the normal variable, and the second variable is the key variable. Since the variables used in training the business prediction basic model are all normal variables without missing values, and the possibility distributions of the samples of these normal variables conform to the normal distribution, the possibility distribution of the generated business prediction possibility sequence based on the samples of the normal variables and the business prediction basic model also conforms to the normal distribution. Therefore, by generating a new sample of the key variable in a way that maps the business prediction possibility sequence to the possibility distribution of the non-randomly missing key variable (that is, performing feature completion processing on the target sample according to the business prediction possibility sequence), the possibility distribution of the newly generated sample of the key variable obtained by the feature completion processing can also conform to the normal distribution (that is, consistent with the normal possibility distribution of the real sample of the key variable). Therefore, the newly generated sample of the key variable obtained by completion makes up for the sample characteristics of the key variable caused by non-random missing, thereby improving the accuracy of business data feature completion. Further, when the non-randomly missing key variable is completed according to the business data feature completion solution of this application and the business risk prediction model is trained on this basis, it is beneficial to improve the prediction accuracy of the trained business risk prediction model, and ultimately beneficial to improving the accuracy of business risk prediction.

[0127] Reference Figure 2 As shown, in some embodiments of this specification, step 101 of obtaining the first variable and the non-randomly missing second variable of the business characteristics in the business data set may include the following steps 201 to 204.

[0128] Step 201. Identify the normal variables and key variables in the business data set.

[0129] In some embodiments of this specification, according to different application scenarios, the business data set may be a set of any suitable business data. For example, in some exemplary embodiments of this specification, taking the prediction of credit risk (default risk) in the financial field as an example, the business data set may be user behavior data (such as transaction behavior, default behavior, credit record, etc.) and user basic data (such as age, occupation, income, assets, marital status, etc.) in different dimensions that affect credit risk prediction. Among them, the user behavior data is the user behavior data of multiple users, and the user basic data is the user basic data of multiple users.

[0130] In some embodiments of this specification, a normal variable in the business dataset refers to a variable in which there is no missing value (or at least no non-random missing value) in the business dataset. That is, a normal variable includes not only non-critical variables without missing values (or at least without non-random missing values), but also critical variables without missing values (or at least without non-random missing values). Among them, a critical variable refers to an important variable that has a relatively large impact on business prediction (i.e., the variable corresponding to the main influencing factor of business prediction); a non-critical variable refers to a secondary variable that has a relatively small impact on business prediction (i.e., the variable corresponding to the secondary influencing factor of business prediction). Correspondingly, the part of the business dataset other than the normal variables has missing values, which are missing variables.

[0131] In some embodiments of this specification, it is possible to identify whether a variable is a normal variable by determining whether there are missing values and / or outliers in the samples of each variable in the business dataset; if a variable has no missing values and / or outliers, then this variable is considered a normal variable, otherwise it is confirmed that this variable is a missing variable.

[0132] Step 202: Confirm the critical variables with data missing in the business dataset.

[0133] In some embodiments of this specification, the secondary variables and critical variables in the business dataset can be pre-specified based on experience. Then, based on the above identification of the normal variables and missing variables in the business dataset, it can be determined which missing variables are critical variables, that is, the critical variables with data missing are identified.

[0134] Step 203: Determine whether the critical variables with data missing belong to random missing. If the critical variables with data missing do not belong to random missing, then execute Step 204; otherwise end (that is, when the critical variables with data missing belong to random missing, sample completion can be performed based on traditional imputation methods).

[0135] Step 204: Confirm that the critical variables with data missing belong to non-random missing.

[0136] In some embodiments of this specification, the samples corresponding to the normal variables in the business dataset can be stratified randomly sampled to divide these normal variables into a training set, a validation set, etc., and a machine learning model can be trained using the training set and the validation set, so as to obtain a basic business prediction model; then, the specified samples (such as a test set, etc.) in the business dataset are input into the basic business prediction model, so as to correspondingly obtain multiple business prediction probability values (that is, a business prediction probability sequence). Among them, the sample of a variable is the variable value of the variable.

[0137] As described in the above text, since the variables used in training the business prediction basic model are all normal variables, and the probability distributions of the samples of these normal variables conform to the normal distribution, the probability distribution of the business prediction probability sequence generated based on the samples of the normal variables and the business prediction basic model also conforms to the normal distribution. This provides favorable conditions for subsequent feature completion of the key variable samples through the business prediction probability sequence (for example, feature completion can be performed using multiple interpolation methods) to obtain new samples of the key variables.

[0138] In some embodiments of this specification, determining the missing type of the key variable with non-random missing may include:

[0139] (1) When the sample observation time point of the key variable exceeds the time range corresponding to the key variable, confirm that the type of the key variable with non-random missing is the first type, and the first type represents non-random missing caused by the retrospective time limit. Wherein, the sample observation time point refers to the retrospective time point (i.e., the specified recent time period); the time range corresponding to the key variable refers to the time span of the stock data of the key variable.

[0140] For example, in an exemplary embodiment of this specification, a certain key variable with non-random missing only stores data within the most recent five years; if the sample observation time point is selected as the data of this key variable in the most recent six years, the data of the key variable in the most recent sixth year will be missing.

[0141] (2) When the data range of the truncation rule of the key variable is smaller than the time range corresponding to the key variable, confirm that the type of the key variable with non-random missing is the second type, and the second type represents non-random missing caused by the truncation rule limit. Wherein, the truncation rule refers to the selection rule for selecting data from the time range corresponding to the key variable.

[0142] For example, in an exemplary embodiment of this specification, a certain key variable with non-random missing stores the annual income data of users within the most recent five years; but the truncation rule is the annual income data of users with an annual income greater than 100,000, then the data of users with an annual income not greater than 100,000 will be missing in the truncated data.

[0143] Furthermore, in some embodiments of this specification, determining the sample selection range of the key variable according to the missing type may include:

[0144] (1) When the type of the key variable with non-random missing is the first type, the sample selection range of the key variable is to select all samples corresponding to the key variable from the business dataset.

[0145] That is, for the non-random missingness caused by the backtracking time limit, when performing multiple interpolation processing subsequently, the sample selection range of this key variable is all samples of this key variable.

[0146] (2) When the type of the key variable with non-random missingness is the second type, the sample selection range of the key variable is to select samples corresponding to the key variable that meet the truncation rule from the business dataset.

[0147] That is, for the non-random missingness caused by the truncation rule limit, when performing multiple interpolation processing subsequently, the sample selection range of this key variable is all samples of this key variable that meet the truncation rule.

[0148] In the embodiments of this specification, based on identifying the non-random missingness type of the key variable, for the non-random missingness caused by the backtracking time limit, the sample selection range is to select all samples corresponding to the key variable from the business dataset; while for the non-random missingness caused by the truncation rule limit, the sample selection range is to select samples corresponding to the key variable that meet the truncation rule from the business dataset. Determining the sample selection range in this way can not only retain the original data as much as possible but also meet the personalized needs of users for sample selection; more importantly, by identifying the non-random missingness type and its sample selection range, it can provide a data basis for constructing different mapping functions for different non-random missingness types to improve the accuracy of feature completion subsequently.

[0149] Reference Figure 3 As shown, in some embodiments of this specification, in step 104, performing feature completion processing on the target sample according to the business prediction possibility sequence may include the following steps 301 to 305.

[0150] Step 301: Perform conversion processing on the target sample to obtain a specified type parameter sequence of the key variable.

[0151] In some embodiments of this specification, the specified type parameter sequence may be an integer value sequence of the key variable. For example, the samples of the key variable within the sample selection range can be mapped to the range of 0 to 100, and the mapped values are rounded to obtain an integer value sequence of the key variable.

[0152] For example, in an exemplary embodiment of this specification, if the key variable with non-random missingness is the user's annual income, and the data of the user's annual income falling within the sample selection range is shown in Table 1 below.

[0153] Table 1

[0154] Name Annual Income (10,000 yuan) Zhang San 12.3 Li Si 15.1 Wang Er 43.8 Zhao Wu 0 Liu Yi 24.9 Yang Qi 78.4 Zhou Ba 102.1 … …

[0155] The annual income of users in Table 1 can be mapped based on the clip function (equivalent to a data normalization process) so that the annual income of users in Table 1 is restricted to the range of 0 to 100, specifically, restricted to the range of [0, 99]. Among them, the clip function is a function in the numpy tool used to limit the element values in an array within a specified range. The clip function will limit the element values in the array between the given minimum and maximum values, for example, limited between [0, 99]. Values outside the range will be mapped to the minimum or maximum value accordingly. After mapping the annual income of users in Table 1 with the clip function and performing rounding on the mapped data, an integer value array of the key variables as shown in Table 2 can be obtained.

[0156] Table 2

[0157] Name Annual Income (10,000 yuan) Zhang San 12 Li Si 15 Wang Er 44 Zhao Wu 0 Liu Yi 25 Yang Qi 78 Zhou Ba 99 … …

[0158] Step 302: Determine the cumulative distribution sequence of the specified type parameter sequence of the key variable.

[0159] In some embodiments of this specification, the cumulative distribution sequence can be a cumulative percentage sequence. Therefore, determining the cumulative distribution sequence of the specified type parameter sequence of the key variable can include:

[0160] (1) Calculate the proportion of each integer value in the integer value sequence of the key variable.

[0161] For example, in an exemplary embodiment of this specification, if the integer value sequence of the key variable includes 10,000 integer values, and among them, there are 100 integer values with a value of 150,000, then the proportion of the integer value with a value of 150,000 is 100÷10,000 = 0.01.

[0162] (2) Calculate the cumulative percentage of the proportion to obtain the cumulative percentage sequence.

[0163] For example, taking the integer value sequence of the annual income of users shown in Table 2 {120,000, 150,000, 440,000, 0, 250,000, 780,000, 990,000,...} as an example, if the corresponding proportions are {r1, r2, r3, r4, r5, r6, r7,...}; then, starting from the first value (120,000) in this sequence, the cumulative percentage calculation process is as follows:

[0164] The cumulative percentage of 120,000 is (r1 / r1)×100%;

[0165] The cumulative percentage of 150,000 is (r2 / (r1+r2))×100%;

[0166] The cumulative percentage of 440,000 is (r3 / (r1+r2+r3))×100%;

[0167] The cumulative percentage of 0 is (r4 / (r1 + r2 + r3 + r4)) × 100%;

[0168] The cumulative percentage after 250,000 is (r5 / (r1 + r2 + r3 + r4 + r5)) × 100%;

[0169] The cumulative percentage of 780,000 is (r6 / (r1 + r2 + r3 + r4 + r5 + r6)) × 100%;

[0170] The cumulative percentage of 990,000 is (r7 / (r1 + r2 + r3 + r4 + r5 + r6 + r7)) × 100%;

[0171] …

[0172] Sorting the calculated cumulative percentages from small to large can obtain the cumulative percentage sequence.

[0173] Step 303: Determine the overall distribution of the service prediction possibility sequence according to the cumulative distribution sequence.

[0174] In some embodiments of this specification, the overall distribution may refer to percentiles; the service prediction possibility sequence is the service prediction probability sequence; therefore, determining the overall distribution of the service prediction possibility sequence according to the cumulative distribution sequence means: determining the percentiles of the service prediction probability sequence according to the cumulative percentage sequence.

[0175] Reference Figure 4 As shown, in some embodiments of this specification, determining the overall distribution of the service prediction possibility sequence according to the cumulative distribution sequence in step 303 may include steps 401 to 403.

[0176] Step 401: Perform a sorting process on the cumulative percentage sequence from small to large.

[0177] Step 402: Perform cumulative distribution complement processing on both ends of the sorted cumulative percentage sequence.

[0178] In some embodiments of this specification, the cumulative distribution complement processing may refer to out-of-bounds value insertion processing.

[0179] Step 403: Convert the cumulative percentage sequence after the cumulative distribution complement processing into the percentiles of the service prediction possibility sequence.

[0180] For example, taking the embodiment shown in Table 2 as an example, since the value types of the integer value sequence of the user's annual income include 0, 10,000, 20,000, …, 990,000, a total of 100, there are also 100 cumulative percentages in the cumulative percentage sequence: {1, a1, a2, a3, a4, a5, a6, …, 0}; when adding the first super boundary value of -1 and adding the second super boundary value of 1.1, after the addition, the cumulative percentage sequence becomes: {0, …, a6, a5, a4, a3, a2, a1, 1}, that is, the percentiles of the business prediction possibility sequence are {0, …, a6, a5, a4, a3, a2, a1, 1}.

[0181] By sorting the cumulative percentage sequence and performing the super boundary value insertion process, the division of percentiles can be facilitated.

[0182] Step 304, generate a mapping relationship according to the overall distribution and the specified type parameter sequence of the key variable.

[0183] In some embodiments of the present specification, the mapping relationship is a mapping function or a mapping method. Refer to Figure 5 As shown, in some embodiments of the present specification, generating the mapping relationship according to the overall distribution and the specified type parameter sequence of the key variable in step 304 may include steps 501 to 503.

[0184] Step 501, perform a segmentation process on the value range of the business prediction possibility sequence according to the overall distribution to obtain a plurality of first value ranges.

[0185] In some embodiments of the present specification, performing a segmentation process on the value range of the business prediction possibility sequence according to the overall distribution may mean: dividing the value range of the business prediction probability array into a plurality of domains according to the percentiles.

[0186] For example, taking the percentiles {0, …, a6, a5, a4, a3, a2, a1, 1} as an example, 0 to 1 can be divided into 99 value intervals, such as (a1, 1], (a2, a1], (a3, a2], (a4, a3], …. Each value interval serves as a domain of the mapping function.

[0187] Step 502, generate a corresponding second value range for each of the first value ranges, where the second value range is an integer falling within the target range.

[0188] In some embodiments of the present specification, generating a corresponding second value range for each of the first value ranges may mean: assigning a corresponding value range to each of the domains, and the value range is an integer falling within 0 to 100.

[0189] For example, taking 99 domains with percentiles {0, …, a6, a5, a4, a3, a2, a1, 1} as an example, the value range can be divided as follows:

[0190] The value range corresponding to (a1, 1] is 99;

[0191] The value range corresponding to (a2, a1] is 98;

[0192] The value range corresponding to (a3, a2] is 97;

[0193] The value range corresponding to (a4, a3] is 96;

[0194] …

[0195] Step 503: For each first value range, determine the union of each first value range and the second value range corresponding to each first value range as the mapping relationship.

[0196] In some embodiments of this specification, determining the union of each first value range and its corresponding second value range as the mapping relationship may mean: determining the union of each domain and its corresponding value range as the mapping function.

[0197] The piecewise function formed by the union of each domain and its corresponding value range is the mapping function, as shown in the following formula:

[0198] Where x is the business prediction probability value and f(x) is the new sample of the key variable.

[0199] Step 305: Input the business prediction possibility sequence into the mapping relationship to obtain the corresponding new sample of the key variable.

[0200] As described above, since the key variables with non-random missing values include two types of non-random missing types, and the sample selection ranges for each non-random missing type are different, therefore, according to different non-random missing types, two mapping functions can be constructed respectively. For example, for the non-random missing caused by the backtracking time limit, the first mapping function f1(x) can be constructed, and for the non-random missing caused by the truncation rule limit, the second mapping function f2(x) can be constructed. In this way, the reliability of the processing result of the feature completion process can be improved.

[0201] For example, if the key variable with non-random missing values is the non-random missing caused by the backtracking time limit, the business prediction probability array can be input into the first mapping function f1(x); if the key variable with non-random missing values is the non-random missing caused by the truncation rule limit, the business prediction probability array can be input into the second mapping function f2(x).

[0202] According to Figure 3 The feature completion processing method in the illustrated embodiment is a new multiple imputation method, which takes into account the uncertainty and variability of missing values, reduces the systematic bias that may be introduced by traditional imputation methods, and makes the probability distribution of the new samples of the key variables obtained by feature completion processing more in line with the normal probability distribution of the real samples of the key variables.

[0203] Figure 6 Fig. shows a schematic diagram of the comparison of the probability density distributions of the real samples and the fitted samples of the key variables with non-random missing values in an exemplary embodiment of this specification. Figure 6 In the figure, the abscissa represents the value, the ordinate represents the proportion of the value, the dark gray histogram is the real sample of the key variable X, the black curve is the probability density distribution of the real sample of the key variable X, the light gray histogram is the new sample of the key variable X obtained by multiple interpolation processing, and the light gray curve is the probability density distribution of the new sample of the key variable X obtained by multiple interpolation processing; it can be seen that the probability density distribution of the new sample of the key variable X obtained by multiple interpolation processing is basically the same as the probability density distribution of the real sample of the key variable X, thus verifying the correctness and feasibility of the business data feature completion method in the embodiments of this specification.

[0204] Figure 7 Fig. shows a schematic diagram of the probability density distribution of all key variables X after multiple interpolation processing of the key variable X with non-random missing values in an exemplary embodiment of this specification; Figure 8a Fig. shows a schematic diagram of the probability density distribution of all key variables X after fitting the key variable X with non-random missing values using a traditional fitting function; Figure 8b Fig. shows a schematic diagram of the probability density distribution of all key variables X after filling the key variable X with non-random missing values using the traditional method of filling missing values with random numbers; Figure 8c Fig. shows a schematic diagram of the probability density distribution of all key variables X after filling the key variable X with non-random missing values using the traditional method of filling missing values with 0. Comparing Figure 7 , Figure 8a , 8b and Figure 8c it can be seen that when filling the key variables with non-random missing values, compared with the traditional methods, the probability density distribution of all key variables X obtained by using the business data feature completion method in the embodiments of this specification is more in line with the normal distribution.

[0205] Corresponding to the above business data feature completion method, this specification also provides a method for training a business risk prediction model, which can be applied to the computer device side. Referring to Figure 9 as shown, in some embodiments of this specification, the method for training a business risk prediction model may include the following steps 901-905.

[0206] Step 901: Obtain the normal variables of business features and the key variables with non-random missing values in the business dataset.

[0207] Step 902: Based on the samples of the normal variables, train a first business risk prediction model, and generate a business risk possibility sequence according to the first business risk prediction model.

[0208] Among them, the initial model of the first business risk prediction model is a machine learning model; the business risk possibility sequence is a sequence composed of business risk probability values.

[0209] Step 903: Determine the missing type of the key variables with non-random missing values, and determine the sample selection range of the key variables according to the missing type.

[0210] Step 904: Perform multiple interpolation processing on the target samples according to the business risk possibility sequence, so that the new samples of the key variables obtained conform to the possibility distribution of the key variables. The target samples are the parameter values of the key variables of the samples within the sample selection range.

[0211] Step 905: Train a second business risk prediction model according to the samples of the normal variables, the new samples of the key variables, and the samples of the key variables with non-random missing values.

[0212] In the embodiments of this specification, since the variables used when training the first business risk prediction model are all normal variables without missing values, and the possibility distributions of the samples of these normal variables conform to the normal distribution, the possibility distribution of the business risk possibility sequence generated based on the samples of the normal variables and this first business risk prediction model also conforms to the normal distribution. Therefore, by mapping the business risk possibility sequence to the possibility distribution mode of the key variables with non-random missing values to generate new samples of the key variables (that is, performing multiple interpolation processing on the samples of the key variables within the sample selection range using the business risk possibility sequence), the possibility distribution of the interpolated new samples of the key variables can also conform to the normal distribution (that is, be consistent with the normal possibility distribution of the true samples of the key variables). Therefore, the interpolated new samples of the key variables can complement the sample characteristics of the key variables caused by non-random missing values, thereby improving the accuracy of business data feature complementation; on this basis, when performing a business risk prediction model according to the samples of the normal variables, the new samples of the key variables, and the samples of the key variables with non-random missing values, the prediction accuracy of the trained business risk prediction model (the second business risk prediction model) can be improved.

[0213] In the method for training a business risk prediction model according to some embodiments of this specification, the multiple interpolation processing of the target sample according to the business risk possibility sequence includes:

[0214] Perform a conversion process on the target sample to obtain a sequence of key variable integer values;

[0215] Determine the cumulative percentage sequence of the sequence of key variable integer values;

[0216] Determine the percentile of the business risk possibility sequence according to the cumulative percentage sequence;

[0217] Generate a mapping relationship according to the percentile and the sequence of key variable integer values;

[0218] Input the business risk possibility sequence into the mapping relationship to obtain corresponding new key variable samples.

[0219] In the method for training a business risk prediction model according to some embodiments of this specification, the determination of the percentile of the business risk possibility sequence according to the cumulative percentage sequence includes:

[0220] Perform a sorting process on the cumulative percentage sequence from small to large;

[0221] Perform an out-of-bounds value insertion process on both ends of the sorted cumulative percentage sequence;

[0222] Convert the cumulative percentage sequence after inserting out-of-bounds values into the percentile of the business risk possibility sequence.

[0223] In the method for training a business risk prediction model according to some embodiments of this specification, the generation of the mapping relationship according to the percentile and the sequence of key variable integer values includes:

[0224] Perform a segmentation process on the value range of the business risk possibility sequence according to the percentile to obtain multiple first value ranges;

[0225] Generate a corresponding second value range for each of the first value ranges, where the second value range is an integer falling within the target range;

[0226] Determine the union of each of the first value ranges and its corresponding second value range as the mapping relationship.

[0227] In the method for training a business risk prediction model according to some embodiments of this specification, the acquisition of normal variables and non-randomly missing key variables in the business dataset includes:

[0228] Identify the normal variables and key variables in the business dataset;

[0229] Identify the key variables with missing data in the business dataset;

[0230] When the key variables with missing data do not belong to random missing, confirm that the key variables with missing data belong to non-random missing.

[0231] In the method for training a business risk prediction model according to some embodiments of this specification, the generating a business risk possibility sequence according to the first business risk prediction model includes:

[0232] Input a part of the samples in the normal variables into the first business risk prediction model to obtain the corresponding business risk possibility sequence.

[0233] In the method for training a business risk prediction model according to some embodiments of this specification, the determining the missing type of the key variables with non-random missing includes:

[0234] When the sample observation time point of the key variable exceeds the time range corresponding to the key variable, confirm that the type of the key variable with non-random missing is the first type, and the first type represents non-random missing caused by retrospective time limit; or,

[0235] When the data range of the truncation rule of the key variable is smaller than the time range corresponding to the key variable, confirm that the type of the key variable with non-random missing is the second type, and the second type represents non-random missing caused by truncation rule limit.

[0236] In the method for training a business risk prediction model according to some embodiments of this specification, the determining the sample selection range of the key variable according to the missing type includes:

[0237] When the type of the key variable with non-random missing is the first type, the sample selection range of the key variable is to select all samples corresponding to the key variable from the business dataset; or,

[0238] When the type of the key variable with non-random missing is the second type, the sample selection range of the key variable is to select the samples corresponding to the key variable that conform to the truncation rule from the business dataset.

[0239] In the embodiments of this specification, Figure 9 For the specific technical details of the business risk prediction model training method shown, reference can be made to the description in the embodiments of the business data feature completion method in the above text, and details will not be repeated.

[0240] Corresponding to the above business risk prediction model training method, this specification also provides a business risk prediction method, which can be applied to the computer device side. Refer toFigure 10 As shown, in some embodiments of this specification, the business risk prediction method may include the following steps 1001 to 1002.

[0241] Step 1001: Obtain the business data of the target object, where the business data includes the normal variable data and the key variable data of the business characteristics of the target object.

[0242] In some embodiments of this specification, the target object may be a user.

[0243] Step 1002: Input the business data into a pre-trained business risk prediction model to obtain the business risk possibility of the target object; the business risk prediction model includes a second business risk prediction model.

[0244] Although the process flow described above includes multiple operations that occur in a specific order, it should be clearly understood that these processes may include more or fewer operations, and these operations may be executed sequentially or in parallel (e.g., using a parallel processor or a multi-threaded environment).

[0245] Corresponding to the above business data feature completion method, an embodiment of this specification also provides a business data feature completion device. Refer to Figure 11 As shown, in some embodiments of this specification, the business data feature completion device may include:

[0246] A first acquisition module 111, configured to acquire the normal variables of the business characteristics and the key variables with non-random missing values in the business data set;

[0247] A first generation module 112, configured to train a business prediction basic model based on the samples of the normal variables, and generate a business prediction possibility sequence according to the business prediction basic model;

[0248] A first determination module 113, configured to determine the missing type of the key variables with non-random missing values, and determine the sample selection range of the key variables according to the missing type;

[0249] A first feature completion module 114, configured to perform multiple interpolation processing on the target sample according to the business prediction possibility sequence, so that the newly obtained key variable samples conform to the possibility distribution of the key variables, where the target sample is the parameter value of the key variables of the samples within the sample selection range.

[0250] Corresponding to the above business risk prediction model training method, an embodiment of this specification also provides a business risk prediction model training device. Refer to Figure 12 As shown, in some embodiments of this specification, the business risk prediction model training device may include:

[0251] A second acquisition module 121, configured to acquire normal variables of business features and key variables with non-random missing values in a business dataset;

[0252] A second generation module 122, configured to train a first business risk prediction model based on samples of the normal variables, and generate a business risk possibility sequence according to the first business risk prediction model;

[0253] A second determination module 123, configured to determine the missing type of the key variables with non-random missing values, and determine the sample selection range of the key variables according to the missing type;

[0254] A second feature completion module 124, configured to perform multiple interpolation processing on a target sample according to the business risk possibility sequence, so that the obtained new samples of the key variables conform to the possibility distribution of the key variables, where the target sample is the parameter value of the key variables of the samples within the sample selection range;

[0255] A training module 125, configured to train a second business risk prediction model according to the samples of the normal variables, the new samples of the key variables, and the samples of the key variables with non-random missing values.

[0256] Corresponding to the above business risk prediction method, an embodiment of this specification further provides a business risk prediction device. Refer to Figure 13 As shown, in some embodiments of this specification, the business risk prediction device may include:

[0257] A third acquisition module 131, configured to acquire business data of a target object, where the business data includes normal variable data and key variable data of the business features of the target object;

[0258] A risk prediction module 132, configured to input the business data into a pre-trained business risk prediction model to obtain the business risk possibility of the target object; the business risk prediction model is the second business risk prediction model obtained according to the above business risk prediction model training method.

[0259] For the convenience of description, when describing the above device, various units are described separately according to functions. Of course, when implementing this specification, the functions of each unit can be implemented in one or more software and / or hardware.

[0260] It should be noted that in the embodiments of this specification, the user information involved (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) are all information and data that have been authorized and consented to by the user and fully authorized by all parties. That is, the acquisition, transmission, storage, use, processing, etc. of data in the technical solution of this application all comply with the relevant provisions of national laws and regulations.

[0261] The embodiments of this specification also provide a computer device. As Figure 14 shown, in some embodiments of this specification, the computer device 1402 may include one or more processors 1404, such as one or more central processing units (CPUs) or graphics processing units (GPUs), and each processing unit may implement one or more hardware threads. The computer device 1402 may also include any memory 1406 for storing any kind of information such as code, settings, data, etc. In a specific embodiment, a computer program stored on the memory 1406 and executable on the processor 1404, when the computer program is run by the processor 1404, may execute the instructions of the business data feature completion, business risk prediction method, or business risk model training method described in any of the above embodiments. Non-limitingly, for example, the memory 1406 may include any one or a combination of the following: any type of RAM, any type of ROM, flash memory devices, hard disks, optical discs, etc. More generally, any memory may use any technology to store information. Further, any memory may provide volatile or non-volatile retention of information. Further, any memory may represent a fixed or removable component of the computer device 1402. In one case, when the processor 1404 executes the associated instructions stored in any memory or combination of memories, the computer device 1402 may perform any operation of the associated instructions. The computer device 1402 also includes one or more drive mechanisms 1408 for interacting with any memory, such as a hard disk drive mechanism, an optical disc drive mechanism, etc.

[0262] The computer device 1402 may also include an input / output interface 1410 (I / O) for receiving various inputs (via the input device 1412) and for providing various outputs (via the output device 1414). A specific output mechanism may include a presentation device 1416 and an associated graphical user interface 1418 (GUI). In other embodiments, the input / output interface 1410 (I / O), the input device 1412, and the output device 1414 may not be included, and it may only be a computer device in a network. The computer device 1402 may also include one or more network interfaces 1420 for exchanging data with other devices via one or more communication links 1422. One or more communication buses 1424 couple the components described above together.

[0263] The communication link 1422 may be implemented in any manner, for example, through a local area network, a wide area network (e.g., the Internet), a point-to-point connection, etc., or any combination thereof. The communication link 1422 may include any combination of hardwired links, wireless links, routers, gateway functions, name servers, etc. governed by any protocol or combination of protocols.

[0264] This application is described with reference to the flowcharts and / or block diagrams of methods, devices (systems), computer-readable storage media, and computer program products according to some embodiments of this specification. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of flows and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processors to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processors generate means for implementing the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0265] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processors to work in a specific manner, so that the instructions stored in the computer-readable memory generate a manufactured article including instruction means, and the instruction means implements the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0266] These computer program instructions can also be loaded onto a computer or other programmable data processors, so that a series of operation steps are executed on the computer or other programmable devices to generate a computer-implemented process. Thus, the instructions executed on the computer or other programmable devices provide means for implementing the functions specified in Figure 1One or more processes and / or blocks Figure 1 Steps of functions specified in one or more blocks

[0267] In a typical configuration, a computer device includes one or more processors (CPUs), an input / output interface, a network interface, and memory.

[0268] Memory may include non-permanent memory in the form of computer-readable media, random access memory (RAM) and / or non-volatile memory such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0269] Computer-readable media includes permanent and non-permanent, removable and non-removable media that can store information by any method or technology. The 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 technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic disk storage or other magnetic storage devices, or any other non-transmission media that can be used to store information accessible by a computer device. As defined in this specification, computer-readable media does not include transitory media such as modulated data signals and carrier waves.

[0270] Those skilled in the art should understand that the embodiments of this specification can be provided as a method, a system, or a computer program product. Therefore, the embodiments of this specification can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the embodiments of this specification can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memory, CD-ROM, optical memory, etc.) containing computer-usable program code.

[0271] Embodiments of this specification can 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. Embodiments of this specification can also be practiced in a distributed computing environment where tasks are performed by remote processors connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage devices.

[0272] It should also be understood that in the embodiments of this specification, the term "and / or" is merely a description of the association relationship between associated objects, indicating that three relationships can exist. For example, A and / or B can represent three situations: A exists alone, A and B exist simultaneously, and B exists alone. Additionally, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship.

[0273] Each embodiment in this specification is described in a progressive manner. For the same or similar parts between the embodiments, reference can be made to each other. Each embodiment focuses on the differences from other embodiments. In particular, for system embodiments, since they are basically similar to method embodiments, the description is relatively simple, and reference can be made to the corresponding parts of the method embodiments for the relevant content.

[0274] In the description of this specification, the description with reference to terms such as "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the embodiments of this specification. In this specification, the schematic descriptions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described can be combined in a suitable manner in any one or more embodiments or examples. In addition, without contradiction, those skilled in the art can combine and combine the different embodiments or examples described in this specification and the features of different embodiments or examples.

[0275] The above are only the embodiments of this application and are not used to limit this application. For those skilled in the art, various changes and modifications can be made to this application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of this application shall be included within the scope of the claims of this application.

Claims

1. A method for completing business data features, characterized in that, Including: Obtaining a first variable of business characteristics and a second variable of non-random missing values in a business dataset; Training a basic business prediction model based on samples of the first variable, and generating a business prediction possibility sequence according to the basic business prediction model; Determining the missing type of the second variable of non-random missing values, and determining the sample selection range of the second variable according to the missing type; Performing feature completion processing on a target sample according to the business prediction possibility sequence, so that the obtained new sample of the second variable conforms to the probability distribution of the second variable, where the target sample is the parameter value of the second variable of the sample within the sample selection range.

2. The method for complementing business data features according to claim 1, wherein The performing feature completion processing on the target sample according to the business prediction possibility sequence includes: Performing conversion processing on the target sample to obtain a specified type parameter sequence of the second variable; Determining the cumulative distribution sequence of the specified type parameter sequence of the second variable; Determining the overall distribution of the business prediction possibility sequence according to the cumulative distribution sequence; Generating a mapping relationship according to the overall distribution and the specified type parameter sequence of the second variable; Inputting the business prediction possibility sequence into the mapping relationship to obtain a corresponding new sample of the second variable.

3. The method for completing business data features according to claim 2, wherein The determining the overall distribution of the business prediction possibility sequence according to the cumulative distribution sequence includes: Performing sorting processing on the cumulative distribution sequence from small to large; Performing cumulative distribution completion processing on both ends of the sorted cumulative distribution sequence; Converting the cumulative distribution sequence after cumulative distribution completion into the overall distribution of the business prediction possibility sequence.

4. The method for complementing business data features according to claim 2, wherein The generating a mapping relationship according to the overall distribution and the specified type parameter sequence of the second variable includes: Performing segmentation processing on the value range of the business prediction possibility sequence according to the overall distribution to obtain a plurality of first value ranges; Generating a corresponding second value range for each of the first value ranges, where the second value range is an integer falling within the target range; For each first value range, determining the union of each first value range and the corresponding second value range of each first value range as the mapping relationship.

5. The method for completing service data features according to claim 1, wherein The generating a business prediction possibility sequence according to the basic business prediction model includes: Inputting a specified part of the samples in the first variable into the basic business prediction model to obtain a corresponding business prediction possibility sequence.

6. The business data feature completion method according to claim 1, wherein The determining the sample selection range of the second variable according to the missing type includes: When the type of the second variable of non-random missing values is the first type, the sample selection range of the second variable is to select all samples corresponding to the second variable from the business dataset; or, When the type of the second variable of non-random missing values is the second type, the sample selection range of the second variable is to select samples corresponding to the second variable that conform to the truncation rule from the business dataset; Wherein, the first type represents non-random missing values caused by retrospective time limit, and the second type represents non-random missing values caused by truncation rule limit.

7. A method for training a business risk prediction model, characterized in that, Including: Obtaining a first variable of business characteristics and a second variable of non-random missing values in a business dataset; Based on the samples of the first variable, a first business risk prediction model is trained, and a business risk possibility sequence is generated according to the first business risk prediction model; Determine the missing type of the second variable with non-random missingness, and determine the sample selection range of the second variable according to the missing type; Perform feature completion processing on the target samples according to the business risk possibility sequence, so that the obtained new samples of the second variable conform to the probability distribution of the second variable, where the target samples are the parameter values of the second variable of the samples within the sample selection range; Based on the samples of the first variable, the new samples of the second variable, and the samples of the second variable with non-random missingness, a second business risk prediction model is trained.

8. The method for training a business risk prediction model according to claim 7, wherein The performing feature completion processing on the target samples according to the business risk possibility sequence includes: Perform conversion processing on the target samples to obtain a specified type parameter sequence of the second variable; Determine the cumulative distribution sequence of the specified type parameter sequence of the second variable; Determine the overall distribution of the business risk possibility sequence according to the cumulative distribution sequence; Generate a mapping relationship according to the overall distribution and the specified type parameter sequence of the second variable; Input the business risk possibility sequence into the mapping relationship to obtain the corresponding new samples of the second variable.

9. The method for training a business risk prediction model according to claim 8, wherein, The determining the overall distribution of the business risk possibility sequence according to the cumulative distribution sequence includes: Perform sorting processing on the cumulative distribution sequence from small to large; Perform cumulative distribution completion processing on both ends of the sorted cumulative distribution sequence; Convert the cumulative distribution sequence after cumulative distribution completion into the overall distribution of the business risk possibility sequence.

10. The method for training a business risk prediction model according to claim 8, wherein, The generating a mapping relationship according to the overall distribution and the specified type parameter sequence of the second variable includes: Perform segmentation processing on the value range of the business risk possibility sequence according to the overall distribution to obtain multiple first value ranges; Generate a corresponding second value range for each of the first value ranges, where the second value range is an integer falling within the target range; For each first value range, determine the union of each first value range and the corresponding second value range of each first value range as the mapping relationship.

11. The method for training a business risk prediction model according to claim 7, wherein The generating a business risk possibility sequence according to the first business risk prediction model includes: Input a part of the samples in the first variable into the first business risk prediction model to obtain the corresponding business risk possibility sequence.

12. The training method of the business risk prediction model according to claim 7, wherein, The determining the sample selection range of the second variable according to the missing type includes: When the type of the second variable with non-random missingness is the first type, the sample selection range of the second variable is to select all the samples corresponding to the second variable from the business dataset; or, When the type of the second variable with non-random missingness is the second type, the sample selection range of the second variable is to select the samples corresponding to the second variable that conform to the truncation rule from the business dataset; Wherein, the first type represents non-random missingness caused by the retrospective time limit, and the second type represents non-random missingness caused by the truncation rule limit.

13. A business risk prediction method, characterized in that, Includes: Obtain the business data of the target object, where the business data includes the first variable data and the second variable data of the business characteristics of the target object; Input the business data into a pre-trained business risk prediction model to obtain the business risk possibility of the target object; the business risk prediction model includes a second business risk prediction model obtained by the method according to any one of claims 7-12.

14. A business data feature completion device, characterized in that, Comprising: A first acquisition module, configured to acquire the first variable and the non-randomly missing second variable of the business characteristics in the business dataset; A first generation module, configured to train a basic business prediction model based on the samples of the first variable, and generate a sequence of business prediction possibilities according to the basic business prediction model; A first determination module, configured to determine the missing type of the non-randomly missing second variable, and determine the sample selection range of the second variable according to the missing type; A first feature completion module, configured to perform feature completion processing on the target sample according to the sequence of business prediction possibilities, so that the obtained new sample of the second variable conforms to the probability distribution of the second variable, where the target sample is the parameter value of the second variable of the sample within the sample selection range.

15. An apparatus for training a business risk prediction model, characterized in that, Comprising: A second acquisition module, configured to acquire the first variable and the non-randomly missing second variable of the business characteristics in the business dataset; A second generation module, configured to train a first business risk prediction model based on the samples of the first variable, and generate a sequence of business risk possibilities according to the first business risk prediction model; A second determination module, configured to determine the missing type of the non-randomly missing second variable, and determine the sample selection range of the second variable according to the missing type; A second feature completion module, configured to perform feature completion processing on the target sample according to the sequence of business risk possibilities, so that the obtained new sample of the second variable conforms to the probability distribution of the second variable, where the target sample is the parameter value of the second variable of the sample within the sample selection range; A training module, configured to train a second business risk prediction model according to the samples of the first variable, the new samples of the second variable, and the samples of the non-randomly missing second variable.

16. A business risk prediction device, characterized in that, Comprising: A third acquisition module, configured to acquire the business data of the target object, where the business data includes the first variable data and the second variable data of the business characteristics of the target object; A risk prediction module, configured to input the business data into a pre-trained business risk prediction model to obtain the business risk possibility of the target object; the business risk prediction model is a second business risk prediction model obtained by the method according to any one of claims 7-12.

17. A computer device, comprising a memory, a processor, and a computer program stored on the memory, characterized in that, When the computer program is run by the processor, it executes the instructions of the method according to any one of claims 1-13.

18. A computer storage medium, on which a computer program is stored, characterized in that, When the computer program is run by the processor of the computer device, it executes the instructions of the method according to any one of claims 1-13.

19. A computer program product, characterized in that, The computer program product includes a computer program, and when the computer program is run by the processor of the computer device, it executes the instructions of the method according to any one of claims 1-13.

Images