Training method of default risk judgment model and default risk judgment method

Abstract

The invention discloses a training method of a default risk judgment model and a default risk judgment method, and relates to the technical field of artificial intelligence and financial science and technology. The method mainly comprises the following steps: performing missing degree analysis on desensitized credit data to obtain high-value features in the desensitized credit data: based on the high-value features, performing coarse classification on the desensitized credit data through a non-neural network model to obtain a training sample of weak supervised learning; and performing weak supervision training on a neural network model according to the training sample to obtain a default risk prediction model. According to the embodiment of the invention, the availability of the desensitized credit data can be improved.

Description

Technical Field

[0001] The present invention relates to the fields of artificial intelligence and financial technology, and in particular to a training method for a default risk judgment model and a default risk judgment method. Background Art

[0002] In many banks, loan business income accounts for the vast majority of their total income, so the loan business is crucial to the economic benefits of the bank. However, the loan business also has certain risks, the most important of which is credit risk, that is, the risk that the customer cannot repay the loan on time. Therefore, banks need to conduct risk assessment when issuing loans to reduce the risk of loan default. Many banks are using machine learning technology to strengthen loan risk control, which can be applied to multiple aspects of loan risk control, including: customer credit assessment, anti-fraud and risk management.

[0003] Machine learning is highly dependent on data. When the data is rational (large data volume, clean features and small deviation), machine learning can allow the model to adaptively learn data features relatively quickly and easily, thereby completing the prediction of corresponding risks. However, due to practical reasons (computing power limitations and the cost of obtaining data) (data needs to be manually labeled, and manual labeling still has accuracy issues, as well as privacy issues), the data obtained often contains missing values ​​and noisy data.

[0004] This problem is particularly prominent in loan risk control scenarios. The personal information data collected for loan defaults involves many dimensions and a wide range of content, as well as some personal privacy issues. Therefore, the collected data may contain many missing data. In addition, a lot of data must be desensitized when collected, which makes it difficult to determine the specific meaning of each feature in the data during machine learning. This further reduces the availability of data and causes trouble for subsequent machine learning and risk prediction. Summary of the invention

[0005] The present invention provides a training method for a default risk judgment model and a default risk judgment method to improve the availability of desensitized credit data.

[0006] According to one aspect of the present invention, a method for training a default risk prediction model is provided, comprising:

[0007] Perform missing degree analysis on the desensitized credit data to obtain high-value features in the desensitized credit data:

[0008] Based on the high-value features, the desensitized credit data is roughly classified by a non-neural network model to obtain training samples for weakly supervised learning;

[0009] The neural network model is weakly supervised trained according to the training samples to obtain a default risk prediction model.

[0010] According to another aspect of the present invention, a method for predicting default risk is provided, comprising:

[0011] Preprocess the credit data of the user to be predicted to obtain the data to be predicted;

[0012] Based on the data to be predicted, obtaining a default risk prediction result of the user to be predicted according to the default risk prediction model;

[0013] Wherein, the default risk prediction model is obtained by training the default risk prediction model according to any embodiment of the present invention.

[0014] According to another aspect of the present invention, there is provided a training device for a default risk prediction model, comprising:

[0015] The analysis module is used to analyze the degree of missingness of the desensitized credit data and obtain high-value features in the desensitized credit data:

[0016] A classification module, used to roughly classify the desensitized credit data based on the high-value features through a non-neural network model to obtain training samples for weakly supervised learning;

[0017] The training module is used to perform weak supervision training on the neural network model according to the training samples to obtain a default risk prediction model.

[0018] According to another aspect of the present invention, there is provided a device for predicting default risk, comprising:

[0019] A preprocessing module, used for preprocessing the credit data of the user to be predicted to obtain the data to be predicted;

[0020] A prediction module, used to obtain a default risk prediction result of the user to be predicted based on the data to be predicted and according to the default risk prediction model;

[0021] Wherein, the default risk prediction model is obtained by training through the training device for the default risk prediction model described in any embodiment of the present invention.

[0022] According to another aspect of the present invention, a computer program product is provided, including a computer program, which, when executed by a processor, implements the training method of the default risk judgment model or the default risk prediction method according to any embodiment of the present invention.

[0023] According to another aspect of the present invention, an electronic device is provided, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the training method of the default risk judgment model or the default risk prediction method described in any embodiment of the present invention.

[0024] According to another aspect of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the training method of the default risk judgment model or the default risk prediction method described in any embodiment of the present invention when executed.

[0025] The embodiment of the present invention successfully converts the desensitized credit data into training samples for weakly supervised training through feature screening and multi-model fusion, thereby improving the availability of the desensitized credit data and reducing the impact of the low quality of the desensitized credit data on subsequent machine learning.

[0026] It should be understood that the contents described in this section are not intended to identify the key or important features of the embodiments of the present invention, nor are they intended to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0028] Figure 1 is a flow chart of a method for training a default risk prediction model provided according to an embodiment of the present invention;

[0029] Figure 2A is a flowchart of a method for training a default risk prediction model according to another embodiment of the present invention;

[0030] Figure 2B is a statistical diagram of missing ratio provided according to another embodiment of the present invention;

[0031] Figure 3 is a flow chart of a method for predicting default risk provided according to an embodiment of the present invention;

[0032] Figure 4is a structural schematic diagram of a training device for a default risk prediction model provided according to another embodiment of the present invention;

[0033] Figure 5 is a schematic structural diagram of a default risk prediction device provided according to another embodiment of the present invention;

[0034] Figure 6 It is a schematic diagram of the structure of an electronic device implementing an embodiment of the present invention. DETAILED DESCRIPTION

[0035] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0036] It should be noted that the terms "first", "second", etc. in the present invention are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products, or devices.

[0037] Figure 1 This is a flowchart of a method for training a default risk prediction model provided by an embodiment of the present invention. This embodiment can be applied to the case where the credit data disclosed by the industry and desensitized is collected and machine learning is performed based on it. The method can be executed by a training device for a default risk prediction model, which can be implemented in the form of hardware and / or software, and can be configured in an electronic device with corresponding data processing capabilities. Figure 1 As shown, the method includes:

[0038] S110. Analyze the degree of missingness of the desensitized credit data to obtain high-value features in the desensitized credit data.

[0039] S120. Based on the high-value features, the desensitized credit data is roughly classified through a non-neural network model to obtain training samples for weakly supervised learning.

[0040] S130. Perform weak supervision training on the neural network model according to the training samples to obtain a default risk prediction model.

[0041] Among them, weakly supervised learning is a machine learning method that uses partially or incompletely labeled data to train models to solve the problem of insufficient labeled data. Unlike traditional supervised learning, weakly supervised learning does not require all data to be labeled, and it can learn useful patterns and knowledge from partially labeled data. In weakly supervised learning, only part of the sample data is labeled, and some data is incompletely supervised without labels; the sample data is only given inaccurate supervision with coarse-grained labels; and inaccurate supervision refers to the labels given are not always correct. Neural networks can be regarded as a complex nonlinear function fitting model. It fits a function by learning the relationship between input and output, so that for a given input, the neural network can predict the corresponding output. Each layer of the neural network can be regarded as a complex nonlinear function. Through the combination of multiple layers of neurons, the neural network can fit very complex functional relationships. It analyzes and processes the input data to solve complex problems by learning to imitate the neural network of the human brain. Neural networks are usually composed of multiple layers of neurons, each layer of neurons is connected by weights, and trained by the back propagation algorithm to minimize the loss function so that the neural network can predict the output results. The reason why neural networks can complete weakly supervised learning is that the trained models are all high-dimensional vectors, and the reason for prediction errors is that the vectors are biased. However, the training data and test data are independent and identically distributed, so we can correct the output vector direction of the test set based on the training in the training set.

[0042] Specifically, we obtain credit data whose key information disclosed by the credit platform has been desensitized. These desensitized credit data usually have multiple features (such as work, bank flow, identity, age, income, address, etc.), but because they have been desensitized, testers can no longer know the actual meaning of each feature. We analyze the features of the desensitized credit data one by one to determine the missing ratio of each feature. For features with a large proportion of missing data, these features will cause certain interference to the subsequent coarse classification. Therefore, we choose to eliminate this part of the features and no longer use them in the coarse classification. We use the remaining features as high-value features, and only use the high-value features of the desensitized credit data in the coarse classification.

[0043] A coarse classification model with a non-neural network structure is selected, and the high-value features in the desensitized credit data are used as the input features of the coarse classification model. The coarse classification model is used to classify each sample of the desensitized credit data to obtain the coarse classification labels of each sample in the desensitized credit data. These coarse classification labels are marked on the corresponding samples to obtain the training samples for weakly supervised learning.

[0044] Select appropriate network nodes as the underlying structure of the neural network model, train the coarse classification labels through training samples with coarse classification labels, and use the test set and validation set to test and verify the neural network. After both the test and validation are passed, the neural network is determined as a default risk prediction model.

[0045] The embodiment of the present invention successfully converts the desensitized credit data into training samples for weakly supervised training through feature screening and multi-model fusion, thereby improving the availability of the desensitized credit data and reducing the impact of the low quality of the desensitized credit data on subsequent machine learning.

[0046] Figure 2A This is a flowchart of a method for training a default risk prediction model provided by another embodiment of the present invention. This embodiment is optimized and improved on the basis of the above embodiment. Figure 2A As shown, the method includes:

[0047] S210. Perform distribution statistics on each feature in the desensitized credit data to obtain missing features with missing values ​​and the missing ratio of each missing feature.

[0048] S220: Determine low-value features among the missing features according to the missing ratio, and determine other features other than the low-value features as high-value features.

[0049] Specifically, for each feature in the desensitized credit data, the distribution feature statistics are performed to check whether there is a particularly large bias in the feature. It can be seen that the distribution of loan data still has a serious left skew. For such features, it may be concluded that the source of these left skews may be because the situations of most customers are relatively consistent. For example, everyone's professional title may not be high, the annual income is low, and most customers have no outstanding credit limits or only a few outstanding credit limits in their credit files. These are more in line with the user portraits of ordinary customers in general routines, so the model needs to focus on obtaining a classification result as accurately as possible for these customers. According to the bias results, the missing features with missing values ​​are further determined and the missing ratio of each missing feature is calculated. The higher the missing ratio, the greater the negative impact of the feature on the subsequent rough classification. A missing ratio threshold can be set to determine the features with a missing ratio lower than the threshold as low-value features, or the missing features can be sorted in order from large to small according to the missing ratio, and the missing features ranked first are determined as low-value features. After determining the low-value features, the remaining features are high-value features. By screening features based on the quantified missing ratio, the accuracy of screening the value of features is improved.

[0050] Optionally, determining the low-value features among the missing features according to the missing ratio includes:

[0051] Draw a missing proportion statistical diagram according to the missing proportion;

[0052] The missing ratio statistical diagram is pushed to the tester, and the low-value features determined by the tester are obtained.

[0053] Specifically, Figure 2B As shown in the figure, a missing ratio statistical diagram can be drawn according to the missing ratio of each missing feature, which shows the features with missing values ​​and the corresponding missing ratios. Testers can manually select some features as low-value features based on actual conditions. By manually participating in the screening of low-value features, the screening accuracy of low-value features can be further improved.

[0054] S230. Using the high-value features as extraction features of desensitized credit data, fine-tuning the non-neural network model through the desensitized credit data to obtain a coarse classification model.

[0055] S240, determining a coarse classification label for each sample in the desensitized credit data according to the coarse classification model; and constructing a weakly supervised training sample according to the coarse classification label for each sample in the desensitized credit data.

[0056] Specifically, the non-neural network model that has been built is selected as the basic model. Then, the non-neural network model is fine-tuned using the desensitized credit data to obtain a rough classification model. There is no need to train the rough classification model from scratch, which greatly improves the efficiency of rough classification.

[0057] Optionally, the number of the coarse classification models is at least two, and determining the coarse classification label of each sample in the desensitized credit data according to the coarse classification model includes:

[0058] For each sample in the desensitized credit data, obtaining the classification result of each coarse classification model on the sample; the classification result includes the probability that the sample belongs to each type;

[0059] The classification results are fused to obtain a rough classification label of the sample.

[0060] Specifically, non-neural network models with different structures are fine-tuned to obtain multiple coarse classification models. For each sample in the desensitized credit data, it is input into different coarse classification models to obtain multiple classification results. These classification results are fused. The fusion method can adopt voting method, averaging method, weighted averaging method, etc. to obtain the final coarse classification label of the sample. By designing and using multiple coarse classification models, overfitting can be reduced, generalization ability can be improved, and classification stability can be enhanced.

[0061] For example, the Extreme Gradient Boosting (XGboost) decision tree model and the Support Vector Machine (SVM) model can be used as non-neural network models in the coarse classification, and greedy parameter adjustment can be used before the coarse classification to iteratively fine-tune the two as much as possible to achieve the best model effect.

[0062] Greedy parameter tuning is an iterative parameter optimization method that adjusts only one parameter at a time while keeping other parameters unchanged to observe the impact of the parameter on model performance. The specific steps are as follows:

[0063] 1. Initialize parameters: First, set an initial value for each parameter.

[0064] 2. Single parameter adjustment: Select a parameter, adjust its value, keep other parameters unchanged, and observe the changes in model performance. Usually, we will try multiple times within the range of possible parameter values ​​to select the parameter value that optimizes model performance.

[0065] 3. Iterative tuning: Repeat step 2 until all parameters have been adjusted once. Then, start tuning again from scratch until the model performance no longer improves significantly.

[0066] The formula for greedy parameter adjustment can be expressed as:

[0067]

[0068] XGBoost, the full name of which is eXtreme Gradient Boosting, is an efficient gradient boosting decision tree algorithm. The training process of the Boosting method is step-by-step, that is, the base models are trained one by one in sequence, the training set of the base models is transformed according to a certain strategy, and then the results predicted by all the base models are linearly integrated to finally generate the final prediction result.

[0069] XGBoost is an additive model composed of k base models. Suppose the tree model we want to train in the tth iteration is f t (x), then:

[0070]

[0071] in, is the prediction result of sample i after the tth iteration, is the prediction result of the first t-1 trees, f t (x i ) is the model (function) of the tth tree. The loss function of XGboost can be expressed as and the true value yi To express:

[0072]

[0073] Here, n is the number of samples.

[0074] We know that the prediction accuracy of the model is determined by the model's bias and variance. The loss function represents the model's bias. If you want to reduce the variance, you need to add a regularization term to the objective function to prevent overfitting. Therefore, the objective function consists of the model's loss function L and the regularization term Ω that suppresses the model's complexity. The objective function is defined as follows:

[0075]

[0076] Here, n is the number of samples.

[0077] Among them, Ω(f i ) is to sum the complexity of all t trees and add it to the objective function as a regularization term to prevent the model from overfitting.

[0078] Support Vector Machine (SVM) is a supervised learning model widely used in classification and regression tasks. In the classification problem, SVM tries to find a hyperplane that maximizes the interval between two categories. This hyperplane is called the maximum margin hyperplane. For our final binary classification problem, where each sample x i There is a label y i ∈-1,1, linearly separable SVM tries to find a hyperplane ω T x+b=0, so that the predicted values ​​of all positive samples are greater than 1, and the predicted values ​​of all negative samples are less than -1. In other words, SVM tries to maximize the following interval:

[0079]

[0080] To find this maximum margin hyperplane, SVM needs to minimize the following loss function:

[0081]

[0082] At the same time, the following constraints are met:

[0083]

[0084] Here ||ω|| 2 is the L2 norm of the weight vector ω, which is used to control the complexity of the model and prevent overfitting. iis to deal with the slack variables introduced by linear inseparability, and C is the penalty parameter used to control the balance between model complexity and classification error.

[0085] S250: Perform weak supervision training on the neural network model according to the training samples to obtain a default risk prediction model.

[0086] The embodiment of the present invention selects a non-neural network model that has been constructed as a basic model, and there is no need to train the coarse classification model from scratch, which greatly improves the efficiency of coarse classification.

[0087] Figure 3 This is a flow chart of a method for predicting default risk provided by an embodiment of the present invention. This embodiment can be applied to the situation where a bank predicts the default risk of a user before lending to the user. The method can be executed by a default risk prediction device, which can be implemented in the form of hardware and / or software, and can be configured in an electronic device with corresponding data processing capabilities. Figure 3 As shown, the method includes:

[0088] S310, pre-processing the credit data of the user to be predicted to obtain the data to be predicted.

[0089] S320: Based on the data to be predicted, obtain a default risk prediction result of the user to be predicted according to the default risk prediction model.

[0090] Wherein, the default risk prediction model is trained by the training method of the default risk prediction model described in any embodiment of the present invention.

[0091] Specifically, by obtaining the credit data of the user to be predicted from different sources (such as bank records, credit reporting agencies, etc.), the credit data is preprocessed as necessary (such as feature selection, standardization / normalization). The preprocessed user data to be predicted is input into the trained default risk prediction model to obtain the default risk prediction result. The prediction result may be the probability of default or a direct binary classification result (default / non-default). If the prediction result is the probability of default, a threshold can be set according to business rules to determine the final prediction category. In addition, the performance of the model needs to be evaluated regularly, and the model parameters should be adjusted or the model should be retrained in a timely manner as market conditions and personal behavior patterns change.

[0092] The embodiment of the present invention predicts the user's default risk before lending by using a pre-trained default risk prediction model, thereby achieving efficient and accurate screening of loan-worthy users and improving the economic benefits of lending.

[0093] On the basis of the above embodiment, optionally, obtaining the default risk prediction result of the user to be predicted based on the data to be predicted and according to the default risk prediction model includes:

[0094] Obtaining preliminary classification results of the data to be predicted through a coarse classification model;

[0095] The data to be predicted and the preliminary classification results are input into a default risk prediction model to obtain a default risk prediction result of the user to be predicted.

[0096] Specifically, model cascading refers to connecting multiple models in series, with the output of the previous model used as the input of the next model to form a multi-stage processing flow. If computing resources and hardware equipment allow, the coarse classification model used in the training process can also be deployed to be cascaded with the default risk prediction model. Before the data to be predicted is input into the default risk prediction model, it is first input into the coarse classification model to obtain the preliminary classification result of the coarse classification model, and then the preliminary classification result and the data to be predicted are input into the default risk prediction model together. By cascading the coarse classification model and the default risk prediction model, the default risk prediction model's prediction ability for default risk is improved.

[0097] Figure 4 A structural diagram of a training device for a default risk prediction model provided by another embodiment of the present invention. Figure 3 As shown, the device comprises:

[0098] The analysis module 410 is used to perform a missing degree analysis on the desensitized credit data to obtain high-value features in the desensitized credit data:

[0099] A classification module 420, configured to roughly classify the desensitized credit data based on the high-value features through a non-neural network model to obtain training samples for weakly supervised learning;

[0100] The training module 430 is used to perform weak supervision training on the neural network model according to the training samples to obtain a default risk prediction model.

[0101] The training device for the default risk prediction model provided in the embodiment of the present invention can execute the training method for the default risk prediction model provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0102] Optionally, the analysis module 410 includes:

[0103] A statistical unit is used to perform distribution statistics on each feature in the desensitized credit data to obtain missing features with missing values ​​and the missing ratio of each missing feature;

[0104] A determination unit is used to determine low-value features among the missing features according to the missing ratio, and determine other features other than the low-value features as high-value features.

[0105] Optionally, the statistical unit is specifically used to: draw a missing ratio statistical diagram according to the missing ratio; push the missing ratio statistical diagram to the tester, and obtain the low-value features determined by the tester

[0106] Optionally, the classification module 420 includes:

[0107] A fine-tuning unit, configured to use the high-value features as extraction features of desensitized credit data, and fine-tune the non-neural network model through the desensitized credit data to obtain a coarse classification model;

[0108] A classification unit, used to determine a coarse classification label of each sample in the desensitized credit data according to the coarse classification model;

[0109] A construction unit is used to construct a weakly supervised training sample according to the coarse classification label of each sample in the desensitized credit data.

[0110] Optionally, the number of the coarse classification models is at least two, and the classification unit is specifically used to: for each sample in the desensitized credit data, obtain the classification results of each coarse classification model for the sample; the classification result includes the probability that the sample belongs to each type; and fuse the classification results to obtain the coarse classification label of the sample.

[0111] The training device for the default risk prediction model further described can also execute the training method for the default risk prediction model provided by any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0112] Figure 5 A schematic diagram of a default risk prediction device provided by another embodiment of the present invention. Figure 5 As shown, the device comprises:

[0113] A preprocessing module 510 is used to preprocess the credit data of the user to be predicted to obtain the data to be predicted;

[0114] Prediction module 520, used for obtaining a default risk prediction result of the user to be predicted based on the data to be predicted and according to the default risk prediction model;

[0115] Wherein, the default risk prediction model is obtained by training using the default risk prediction model training device described in any embodiment of the present invention.

[0116] The default risk prediction device provided in the embodiment of the present invention can execute the default risk prediction method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0117] Optionally, the prediction module 520 includes:

[0118] A preliminary classification unit, used for obtaining a preliminary classification result of the data to be predicted through a coarse classification model;

[0119] The prediction unit is used to input the data to be predicted and the preliminary classification result into the default risk prediction model to obtain the default risk prediction result of the user to be predicted.

[0120] The default risk prediction device further described can also execute the default risk prediction method provided by any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0121] Figure 6 A schematic diagram of the structure of an electronic device 60 that can be used to implement an embodiment of the present invention is shown. The electronic device is 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. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.) and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or required herein.

[0122] like Figure 6 As shown, the electronic device 60 includes at least one processor 61, and a memory connected to the at least one processor 61, such as a read-only memory (ROM) 62, a random access memory (RAM) 63, etc., wherein the memory stores a computer program that can be executed by at least one processor, and the processor 61 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 62 or the computer program loaded from the storage unit 68 to the random access memory (RAM) 63. In the RAM 63, various programs and data required for the operation of the electronic device 60 can also be stored. The processor 61, the ROM 62, and the RAM 63 are connected to each other via a bus 64. An input / output (I / O) interface 65 is also connected to the bus 66.

[0123] A number of components in the electronic device 60 are connected to the I / O interface 65, including: an input unit 66, such as a keyboard, a mouse, etc.; an output unit 67, such as various types of displays, speakers, etc.; a storage unit 68, such as a disk, an optical disk, etc.; and a communication unit 69, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 69 allows the electronic device 60 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0124] The processor 61 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the processor 61 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The processor 61 executes the various methods and processes described above, such as a training method for a default risk judgment model or a method for predicting default risk.

[0125] In some embodiments, the training method of the default risk judgment model or the prediction method of the default risk may be implemented as a computer program, which is tangibly contained in a computer-readable storage medium, such as a storage unit 68. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 60 via the ROM 62 and / or the communication unit 69. When the computer program is loaded into the RAM 63 and executed by the processor 61, one or more steps of the training method of the default risk judgment model or the prediction method of the default risk described above may be executed. Alternatively, in other embodiments, the processor 61 may be configured to execute the training method of the default risk judgment model or the prediction method of the default risk by any other appropriate means (e.g., by means of firmware).

[0126] Various implementations 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 chips (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0127] Computer programs for implementing the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that when the computer program is executed by the processor, the functions / operations specified in the flow chart and / or block diagram are implemented. The computer program may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0128] In the context of the present invention, a computer-readable storage medium may be a tangible medium that may contain or store a computer program for use by or in combination with an instruction execution system, device or equipment. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0129] To provide interaction with a user, the systems and techniques described herein may be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device. Other types of devices may also be used to provide interaction with the user; for example, the feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form (including acoustic input, voice input, or tactile input).

[0130] The systems and techniques described herein may be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0131] A computing system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The client and server relationship is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services.

[0132] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps described in the present invention can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solution of the present invention can be achieved, and this document does not limit this.

[0133] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A training method for a default risk prediction model, characterized in that: The method comprises: Perform missing degree analysis on the desensitized credit data to obtain high-value features in the desensitized credit data: Based on the high-value features, the desensitized credit data is roughly classified by a non-neural network model to obtain training samples for weakly supervised learning; The neural network model is weakly supervised trained according to the training samples to obtain a default risk prediction model.

2. The method according to claim 1, characterized in that The high-value features in the desensitized credit data obtained by performing a missing degree analysis on the desensitized credit data include: Perform distribution statistics on each feature in the desensitized credit data to obtain missing features with missing values ​​and the missing ratio of each missing feature; Low-value features among the missing features are determined according to the missing ratio, and other features other than the low-value features are determined as high-value features.

3. The method according to claim 2, characterized in that Determining the low-value features among the missing features according to the missing ratio comprises: Draw a missing proportion statistical diagram according to the missing proportion; The missing ratio statistical diagram is pushed to the tester, and the low-value features determined by the tester are obtained.

4. The method according to claim 1, characterized in that: Based on the high-value features, the desensitized credit data is roughly classified by a non-neural network model to obtain training samples for weakly supervised learning, including: Using the high-value features as extraction features of desensitized credit data, and fine-tuning the non-neural network model through the desensitized credit data to obtain a coarse classification model; Determine a coarse classification label for each sample in the desensitized credit data according to the coarse classification model; A weakly supervised training sample is constructed according to the coarse classification label of each sample in the desensitized credit data.

5. The method according to claim 4, characterized in that The number of the coarse classification models is at least two, and determining the coarse classification label of each sample in the desensitized credit data according to the coarse classification model includes: For each sample in the desensitized credit data, obtaining the classification result of each coarse classification model on the sample; the classification result includes the probability that the sample belongs to each type; The classification results are fused to obtain a rough classification label of the sample.

6. A method for predicting default risk, characterized in that: The method comprises: Preprocess the credit data of the user to be predicted to obtain the data to be predicted; Based on the data to be predicted, obtaining a default risk prediction result of the user to be predicted according to the default risk prediction model; Wherein, the default risk prediction model is obtained by training through the method described in any one of claims 1-5.

7. The method according to claim 6, characterized in that The step of obtaining the default risk prediction result of the user to be predicted based on the data to be predicted and according to the default risk prediction model comprises: Obtaining preliminary classification results of the data to be predicted through a coarse classification model; The data to be predicted and the preliminary classification results are input into a default risk prediction model to obtain a default risk prediction result of the user to be predicted.

8. A training device for a default risk prediction model, characterized in that: The device comprises: An analysis module, used to analyze the degree of missingness of the desensitized credit data and obtain high-value features in the desensitized credit data; A classification module, used to roughly classify the desensitized credit data based on the high-value features through a non-neural network model to obtain training samples for weakly supervised learning; The training module is used to perform weak supervision training on the neural network model according to the training samples to obtain a default risk prediction model.

9. A default risk prediction device, characterized in that: The device comprises: A preprocessing module, used for preprocessing the credit data of the user to be predicted to obtain the data to be predicted; A prediction module, used to obtain a default risk prediction result of the user to be predicted based on the data to be predicted and according to the default risk prediction model; Wherein, the default risk prediction model is obtained by training the device described in claim 8.

10. An electronic device, characterized in that: The electronic device comprises: at least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the training method of the default risk judgment model described in any one of claims 1-5 or the default risk prediction method described in any one of claims 6-7.

11. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the training method of the default risk judgment model described in any one of claims 1 to 5 or the default risk prediction method described in any one of claims 6 to 7 when executed.

Images