Neural evolution-based credit scoring method and device, and electronic equipment

Abstract

The invention discloses a credit scoring method and device based on neural evolution, and electronic equipment, and relates to the field of artificial intelligence, and the credit scoring method comprises the steps: obtaining a target credit scoring model constructed based on a neural evolution strategy, obtaining the credit data of a target object, processing the credit data, obtaining the target credit data, and storing the target credit data in a database; and inputting the target credit data into the target credit scoring model, and outputting a scoring result of the target object. According to the invention, the technical problem of low accuracy of credit scoring of clients in the prior art is solved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence, and in particular to a credit scoring method based on neural evolution, a device thereof, and an electronic device. Background Art

[0002] Credit scoring refers to the process of evaluating an individual or institution's credit history, financial status, repayment ability, and other aspects to give a score or grade that reflects its credit risk level. This score or grade can help creditors such as financial institutions determine the possibility of a borrower repaying its debts, thereby deciding whether to issue loans, credit cards or other credit products to it, as well as determining the interest rate and loan amount. The functions of credit scoring include: (1) Risk assessment: Through credit scoring, creditors can more accurately assess the credit risk level of borrowers. A high credit score usually means that the borrower has good credit and is more likely to repay the debt; a low credit score may mean that the borrower has poor credit and is at risk of default. (2) Determining credit limits: Based on credit scores, creditors can decide the size of the credit limit to be provided to borrowers. Borrowers with high credit scores can usually obtain higher credit limits, while those with low credit scores may be restricted. (3) Determining interest rates: Credit scores can also affect the interest rates that borrowers receive when they obtain loans. Borrowers with high credit scores may enjoy lower interest rates because creditors believe that they have a lower risk of repaying their debts. (4) Reducing risks: Through credit scoring, creditors can better manage risks, avoid lending to high-risk borrowers, reduce default risks, and protect their own interests. In short, credit scores play an important role in the financial field, helping creditors make wise credit decisions, balance risks and returns, and protect the safety of funds. At the same time, for borrowers, a good credit score can also bring more credit opportunities and more favorable credit conditions.

[0003] Currently, since neural networks have the advantages of processing complex nonlinear relationships, extracting high-order features, high prediction accuracy, automatic processing and high timeliness, a credit scoring model can be built based on neural networks to achieve credit scoring for customers.

[0004] In the related art, the back propagation algorithm is generally used to train the neural network to adjust the weights in the neural network and minimize the loss function. Although the back propagation algorithm can adjust the weights of the neural network more efficiently and quickly, so that it gradually converges to the optimal solution, it can also adapt to different complex patterns and data distributions, and various types of neural network structures, including: deep neural networks and convolutional neural networks. However, the use of the back propagation algorithm to train the neural network still has the following problems:

[0005] (1) Local optimal solution: Back propagation may fall into a local optimal solution instead of a global optimal solution. This may cause the model performance to reach a local optimal level instead of a global optimal level.

[0006] (2) Gradient vanishing or gradient exploding: In deep neural networks, backpropagation may face the problem of gradient vanishing or gradient exploding, leading to training difficulties.

[0007] (3) Overfitting: Backpropagation can easily lead to overfitting, especially when the amount of training data is small or the model complexity is high.

[0008] (4) Hyperparameter sensitivity: The choice of hyperparameters such as learning rate and batch size in back-propagation has a significant impact on the training results and needs to be carefully adjusted.

[0009] (5) It is usually necessary to manually design the network structure, activation function and other information. Even for the most experienced experts, it is difficult to ensure that the designed network is the most effective and concise.

[0010] To the above-mentioned problems, no effective solution has been proposed yet. Summary of the invention

[0011] The embodiments of the present invention provide a credit scoring method based on neural evolution, a device thereof, and an electronic device, so as to at least solve the technical problem of low accuracy in credit scoring of customers in the related art.

[0012] According to one aspect of an embodiment of the present invention, a credit scoring method based on neuroevolution is provided, including: obtaining a target credit scoring model constructed based on a neuroevolution strategy; obtaining credit data of a target object, and processing the credit data to obtain target credit data; inputting the target credit data into a target credit scoring model, and outputting a scoring result for the target object.

[0013] Furthermore, before obtaining the target credit scoring model constructed based on the neuroevolution strategy, it also includes: constructing an initial credit scoring model, wherein the model structure of the initial credit scoring model includes at least: a multi-layer neural network layer, the neural network layer includes at least: an input layer, a fully connected layer, and an output layer, each neural network layer includes: a plurality of neurons, and there are corresponding weights between the neurons in each neural network layer and the neurons in the adjacent neural network layer, and the adjacent neural network layer refers to the previous neural network layer or the next neural network layer of the neural network layer; collecting historical credit data and processing the historical credit data to obtain training data; based on the training data, using the neuroevolution strategy to evolve the initial credit scoring model to obtain the target credit scoring model.

[0014] Furthermore, the step of processing the historical credit data to obtain training data includes: cleaning the historical credit data to obtain cleaned historical credit data; marking each historical credit record in the cleaned historical credit data to obtain a credit label corresponding to each historical credit record, wherein the credit label includes: a normal label or a risk label; based on the credit label corresponding to each historical credit record, screening the cleaned historical credit data to obtain target historical credit data, wherein the number of historical credit records corresponding to the normal labels contained in the target historical credit data is equal to the number of historical credit records corresponding to the risk labels; and standardizing the target historical credit data to obtain training data.

[0015] Furthermore, based on the training data, the initial credit scoring model is evolved using a neural evolution strategy to obtain a target credit scoring model, including: determining a preset number of individuals to obtain an initialized population, wherein each individual represents an initial credit scoring model; encoding each individual to obtain a neural network gene corresponding to the individual, wherein the neural network gene includes: multiple node genes and connection information corresponding to each node gene; evolving each neural network gene to obtain an evolved neural network gene; based on the training data, determining a first fitness value of each evolved neural network gene; selecting an evolved neural network gene corresponding to a first fitness value greater than a preset fitness threshold, and evolving the selected evolved neural network gene until the number of evolutions reaches a preset number of evolutions, stopping the evolution, and obtaining a preset neural network gene; based on the training data, determining a second fitness value of each preset neural network gene, and determining the preset neural network gene corresponding to the maximum second fitness value as the target neural network gene; based on the target neural network gene, determining a target credit scoring model.

[0016] Furthermore, the step of encoding each individual to obtain the neural network gene corresponding to the individual includes: determining the coordinate parameters of each neuron in the initial credit scoring model represented by the individual, wherein the coordinate parameters include: layer parameters and bit parameters; combining all neurons in pairs to obtain multiple pairs of neurons; inputting the coordinate parameters of the two neurons corresponding to each pair of neurons into the composite pattern generation network to obtain the weight between the two neurons; characterizing each neuron as a node gene, and characterizing the weight between the neuron and another neuron as connection information; determining the neural network gene based on all node genes and the connection information corresponding to each node gene.

[0017] Furthermore, the step of evolving each neural network gene to obtain an evolved neural network gene includes: performing crossover evolution on every two neural network genes according to a first preset probability; after performing crossover evolution on all neural network genes, performing mutation evolution on each neural network gene according to a second preset probability to obtain an evolved neural network gene.

[0018] Furthermore, based on the training data, the step of determining the first fitness value of each evolved neural network gene includes: converting each evolved neural network gene into an evolved credit scoring model; using the evolved credit scoring model to process each historical credit record in the training data to obtain a predicted scoring result corresponding to the historical credit record, wherein the predicted scoring result includes: a normal credit record or a risky credit record; determining an index value of the evolved credit scoring model based on the predicted scoring result corresponding to each historical credit record and a pre-labeled credit label; and determining the index value as the first fitness value of the evolved neural network gene.

[0019] According to another aspect of an embodiment of the present invention, a credit scoring device based on neuroevolution is also provided, including: an acquisition unit, used to acquire a target credit scoring model constructed based on a neuroevolution strategy; a processing unit, used to acquire credit data of a target object and process the credit data to obtain target credit data; an input unit, used to input the target credit data into the target credit scoring model and output a scoring result for the target object.

[0020] Furthermore, the credit scoring device also includes: a construction unit, which is used to construct an initial credit scoring model before obtaining a target credit scoring model constructed based on a neural evolution strategy, wherein the model structure of the initial credit scoring model includes at least: a multi-layer neural network layer, the neural network layer includes at least: an input layer, a fully connected layer, and an output layer, each neural network layer includes: a plurality of neurons, and there are corresponding weights between neurons in each neural network layer and neurons in an adjacent neural network layer, and an adjacent neural network layer refers to an upper neural network layer or a lower neural network layer of the neural network layer; a collection unit, which is used to collect historical credit data and process the historical credit data to obtain training data; an evolution unit, which is used to evolve the initial credit scoring model based on the training data using a neural evolution strategy to obtain a target credit scoring model.

[0021] Furthermore, the collection unit includes: a first cleaning module, which is used to clean the historical credit data to obtain the cleaned historical credit data; a first labeling module, which is used to label each historical credit record in the cleaned historical credit data to obtain a credit label corresponding to each historical credit record, wherein the credit label includes: a normal label or a risk label; a first screening module, which is used to screen the cleaned historical credit data based on the credit label corresponding to each historical credit record to obtain target historical credit data, wherein the number of historical credit records corresponding to the normal label contained in the target historical credit data is equal to the number of historical credit records corresponding to the risk label; a first processing module, which is used to perform standardization processing on the target historical credit data to obtain training data.

[0022] Furthermore, the evolution unit includes: a first determination module, which is used to determine a preset number of individuals to obtain an initialized population, wherein each individual represents an initial credit scoring model; a first encoding module, which is used to encode each individual to obtain a neural network gene corresponding to the individual, wherein the neural network gene includes: multiple node genes and connection information corresponding to each node gene; a first evolution module, which is used to evolve each neural network gene to obtain an evolved neural network gene; a second determination module, which is used to determine the first fitness value of each evolved neural network gene based on training data; a first selection module, which is used to select an evolved neural network gene corresponding to a first fitness value greater than a preset fitness threshold, and evolve the selected evolved neural network gene until the number of evolutions reaches a preset number of evolutions, stops the evolution, and obtains the preset neural network gene; a third determination module, which is used to determine the second fitness value of each preset neural network gene based on the training data, and determines the preset neural network gene corresponding to the maximum second fitness value as the target neural network gene; a fourth determination module, which is used to determine the target credit scoring model based on the target neural network gene.

[0023] Furthermore, the first encoding module includes: a first determination submodule, used to determine the coordinate parameters of each neuron in the initial credit scoring model of the individual representative, wherein the coordinate parameters include: layer parameters and bit parameters; a first combination submodule, used to combine all neurons in pairs to obtain multiple pairs of neuron pairs; a first input submodule, used to input the coordinate parameters of the two neurons corresponding to each pair of neurons into the composite pattern generation network to obtain the weight between the two neurons; a first characterization submodule, used to characterize each neuron as a node gene, and characterize the weight between the neuron and another neuron as connection information; a second determination submodule, used to determine the neural network gene based on all node genes and the connection information corresponding to each node gene.

[0024] Furthermore, the first evolution module includes: a first evolution submodule, which is used to perform cross-evolution on every two neural network genes according to a first preset probability; and a second evolution submodule, which is used to perform mutation evolution on each neural network gene according to a second preset probability after cross-evolution on all neural network genes to obtain evolved neural network genes.

[0025] Furthermore, the second determination module includes: a first conversion submodule, used to convert each evolved neural network gene into an evolved credit scoring model; a first processing submodule, used to use the evolved credit scoring model to process each historical credit record in the training data to obtain a predicted scoring result corresponding to the historical credit record, wherein the predicted scoring result includes: a normal credit record or a risky credit record; a third determination submodule, used to determine the index value of the evolved credit scoring model based on the predicted scoring result corresponding to each historical credit record and a pre-marked credit label; and a fourth determination submodule, used to determine the index value as the first fitness value of the evolved neural network gene.

[0026] According to another aspect of an embodiment of the present invention, a computer program product is also provided, including a non-volatile computer-readable storage medium, wherein the non-volatile computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, any one of the above-mentioned neuroevolution-based credit scoring methods is implemented.

[0027] According to another aspect of an embodiment of the present invention, there is also provided an electronic device, comprising one or more processors and a memory, wherein the memory is used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement any one of the above-mentioned neuroevolution-based credit scoring methods.

[0028] In the present invention, by obtaining a target credit scoring model constructed based on a neuroevolutionary strategy, obtaining the credit data of the target object, and processing the credit data to obtain the target credit data, the target credit data is input into the target credit scoring model, and the scoring result for the target object is output, thereby solving the technical problem of low accuracy in credit scoring of customers in related technologies.

[0029] In the present invention, a target credit scoring model can be obtained by evolving a neural network through a neuroevolution strategy, and then the credit data of a target object is processed through the target credit scoring model to obtain a scoring result of the target object, thereby achieving the technical effect of efficiently and accurately constructing a credit scoring model, and accurately and quickly evaluating a customer's credit score through a credit scoring model. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0031] Figure 1 A hardware structure block diagram of a computer terminal (or mobile device) for implementing a credit scoring method based on neuroevolution is shown;

[0032] Figure 2 is a flowchart of a credit scoring method based on neuroevolution according to Example 1 of the present invention;

[0033] Figure 3 is a schematic diagram of a neuron structure according to Example 1 of the present invention;

[0034] Figure 4 is a schematic diagram of neural network encoding according to embodiment 1 of the present invention;

[0035] Figure 5 is a schematic diagram of an optional neural evolution-based credit scoring device according to an embodiment of the present invention;

[0036] Figure 6 is a structural block diagram of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0037] 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.

[0038] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings 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.

[0039] It should be noted that the credit scoring method and device based on neuroevolution in the present invention can be used in the field of artificial intelligence (when credit scoring is based on neuroevolution), and can also be used in any field other than the field of artificial intelligence (when credit scoring is based on neuroevolution). The present invention does not limit the application field of the credit scoring method and device based on neuroevolution.

[0040] It should be noted that the relevant information (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.) collected and involved in the present invention are all information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of relevant data are in compliance with the relevant laws, regulations and standards of the relevant regions, necessary confidentiality measures are taken, and public order and good customs are not violated, and corresponding operation entrances are provided for users to choose to authorize or refuse. For example, an interface is set between the system and the relevant users or organizations. Before obtaining relevant information, it is necessary to send an acquisition request to the aforementioned user or organization through the interface, and obtain relevant information after receiving the consent information fed back by the aforementioned user or organization.

[0041] In the present invention, since the neuroevolution algorithm has the advantages of effectively coping with complex environments, automatic parameter adjustment, exploring a broader solution space, overcoming local optimal solutions, being applicable to non-convex optimization problems and parallel training, the introduction of the neuroevolution algorithm can bring new possibilities and efficiency improvements to the training and optimization of neural networks.

[0042] In the present invention, by applying the neuroevolution algorithm to credit scoring, the credit scoring model is trained in an evolutionary manner, which can not only adjust the weights of the neural network, but also evolve the structure of the neural network, thereby achieving the following beneficial effects:

[0043] (1) Improved the accuracy of the credit scoring model in credit assessment.

[0044] (2) Due to the adjustability of the model, it is possible to continuously track the information of the assessment object, obtain new data, and continuously adjust the evolution model, thereby improving the real-time nature of the credit score and saving the cost of re-assessment.

[0045] (3) It reduces manual intervention in model weight and structure training, further reducing the design cost of the model.

[0046] (4) For different detailed businesses, migration can be carried out based on the original structure, and then adjustments can be made with new data to adapt to changes in demand.

[0047] The present invention is described in detail below in conjunction with various embodiments.

[0048] Example 1

[0049] According to an embodiment of the present application, an embodiment of a credit scoring method based on neuroevolution is also provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0050] The method embodiment provided in the first embodiment of the present application can be executed in a mobile terminal, a computer terminal or a similar computing device. Figure 1 FIG. 1 shows a hardware structure block diagram of a computer terminal (or mobile device) for implementing a credit scoring method based on neural evolution. Figure 1 As shown, the computer terminal 10 (or mobile device) may include one or more ( Figure 1 102a, 102b, ..., 102n are used to illustrate) processor 102 (processor 102 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory 104 for storing data, and a transmission device 106 for communication functions. In addition, it may also include: a display, a keyboard, a cursor control device, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the BUS bus), a network interface, a power supply and / or a camera, wherein the network interface may be connected to a wired and / or wireless network. A person of ordinary skill in the art will understand that Figure 1 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 1 More or fewer components as shown, or with Figure 1 Different configurations are shown.

[0051] It should be noted that the one or more processors 102 and / or other data processing circuits described above may generally be referred to herein as "data processing circuits". The data processing circuits may be embodied in whole or in part as software, hardware, firmware, or any other combination thereof. In addition, the data processing circuit may be a single independent processing module, or may be incorporated in whole or in part into any of the other components in the computer terminal 10 (or mobile device). As described in the embodiments of the present application, the data processing circuit acts as a processor control (e.g., selection of a variable resistor terminal path connected to an interface).

[0052] The memory 104 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the credit scoring method based on neuroevolution in the embodiment of the present application. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, that is, realizing the above-mentioned credit scoring method based on neuroevolution. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 104 may further include a memory remotely arranged relative to the processor 102, and these remote memories may be connected to the computer terminal 10 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0053] The transmission device 106 is used to receive or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of the computer terminal 10. In one example, the transmission device 106 includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 106 can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0054] The display may be, for example, a touch screen liquid crystal display (LCD) that enables a user to interact with a user interface of the computer terminal 10 (or mobile device).

[0055] Under the above operating environment, this application provides Figure 2 The neuroevolution-based credit scoring method shown. Figure 2 is a flowchart of a credit scoring method based on neural evolution according to Example 1 of the present invention, such as Figure 2 As shown, the method comprises the following steps:

[0056] Step S201, obtaining a target credit scoring model constructed based on a neuroevolution strategy.

[0057] In an embodiment of the present invention, by introducing a neuroevolution strategy, a credit scoring model can be constructed and optimized. In this way, not only the weight of the model (i.e., the connection strength between neurons) can be adjusted, but also the structure of the model (such as the number and connection mode of neurons) can be dynamically improved through selection, crossover and mutation operations in neuroevolution strategies (such as genetic algorithms).

[0058] In some optional embodiments, the neuroevolution strategy includes the following processes: population initialization: that is, creating a population consisting of multiple neural network structures, each neural network is an individual; fitness evaluation: that is, using a defined fitness function (such as accuracy, recall, F1 score (i.e., an evaluation indicator of comprehensive precision and recall), etc.) to evaluate the performance of each individual (neural network) on historical credit data; genetic operation: that is, selection based on the fitness of the individual (retaining the best performing individual), crossover (combining the advantages of two individuals to create a new individual) and mutation (randomly changing the structure or weight of an individual to increase population diversity); evolutionary iteration: that is, repeating fitness evaluation and genetic operations until a predetermined stopping condition is reached, such as the maximum number of iterations or the fitness reaches a certain level.

[0059] In the embodiment of the present invention, the target credit scoring model constructed by the neuroevolution strategy can automatically learn and adjust its structure and weights to more accurately score credit risk. Then, by obtaining the target credit scoring model constructed based on the neuroevolution strategy, it is convenient to subsequently perform credit scoring on the target object according to the target credit scoring model.

[0060] Step S202, obtaining the credit data of the target object, and processing the credit data to obtain target credit data.

[0061] In the embodiment of the present invention, the target object refers to an individual or institutional client who needs to be scored.

[0062] In an embodiment of the present invention, the credit data of the target object can be collected from the record system of the financial institution, credit reports and other relevant sources, including but not limited to: customer information data, social media data, behavioral data, etc., wherein the customer information data includes: the target object's loan record, repayment history, credit card bills, income status, etc.; the social media data includes: the target object's behavior and relationship data on social media; the behavioral data includes: the target object's online shopping records, mobile payment records, etc.

[0063] The collected credit data can then be processed to obtain target credit data to improve data quality, reduce the impact of noise on the model, and match the data format with model requirements.

[0064] In an embodiment of the present invention, processing the collected credit data may include the following steps:

[0065] Data cleaning: Remove duplicate data, handle missing values ​​and outliers to ensure data quality and integrity.

[0066] Feature standardization / normalization: Scale the feature data to the same range to prevent the model from being affected by different feature scales.

[0067] Feature selection: Screen out features that have a significant impact on the target variable to reduce model complexity and improve model efficiency.

[0068] Feature fusion: Use domain knowledge to build new features to improve the expressiveness and generalization ability of the model.

[0069] Step S203: input the target credit data into the target credit scoring model, and output the scoring result for the target object.

[0070] In an embodiment of the present invention, the processed target credit data may be passed as input to a credit scoring model constructed by a neuroevolution strategy, thereby obtaining a scoring result for the target object.

[0071] In the embodiment of the present invention, when data is input, each eigenvalue corresponds to one or more input nodes in the neural network. The neurons in the model process the input data according to their structure and weights. The data passes through each layer of neurons and is calculated by the activation function until the output layer of the model, generating a numerical result. The result of the output layer represents the credit score of the target object, which is usually a numerical value reflecting the credit risk level, such as a credit score or a probability of default.

[0072] In the embodiment of the present invention, a model constructed by a neuroevolution strategy is used to convert the credit data of the target object into a scoring result, which will guide the decision-making of financial institutions in loan approval, credit line allocation and risk control. Through the neuroevolution model, the scoring result can more accurately reflect the credit status of the target object, thereby helping financial institutions make more reasonable credit decisions.

[0073] In summary, by evolving the neural network through the neuroevolution strategy, the target credit scoring model can be obtained, and then the credit data of the target object is processed through the target credit scoring model to obtain the scoring result of the target object, thereby achieving the technical effect of efficiently and accurately constructing a credit scoring model, and accurately and quickly evaluating the credit score of the customer through the credit scoring model, thereby solving the technical problem of low accuracy in credit scoring of customers in related technologies.

[0074] In order to improve the accuracy of the training target credit scoring model, in the credit scoring method based on neuroevolution provided in Example 1 of the present application, before obtaining the target credit scoring model constructed based on the neuroevolution strategy, an initial credit scoring model is constructed, wherein the model structure of the initial credit scoring model includes at least: a multi-layer neural network layer, the neural network layer includes at least: an input layer, a fully connected layer, and an output layer, each neural network layer includes: a plurality of neurons, and there are corresponding weights between the neurons in each neural network layer and the neurons in the adjacent neural network layer, and the adjacent neural network layer refers to the previous neural network layer or the next neural network layer of the neural network layer; historical credit data is collected and processed to obtain training data; based on the training data, the initial credit scoring model is evolved using the neuroevolution strategy to obtain the target credit scoring model.

[0075] In an embodiment of the present invention, an initial credit scoring model may be constructed first. This model is based on a multi-layer neural network design, and its structure includes at least: a multi-layer neural network layer, namely an input layer, one or more fully connected layers, and an output layer. Each neural network layer is composed of multiple neurons, which are responsible for processing specific input features or transmitting intermediate results. In a neural network, neurons in each neural network layer have connection weights with neurons in an adjacent neural network layer (i.e., the previous neural network layer or the next neural network layer of the neural network layer), i.e., the weight values ​​connecting the neurons between the two layers. These weights represent the intensity of the influence of the input features on the final scoring results.

[0076] Figure 3 is a schematic diagram of a neuron structure according to Example 1 of the present invention, such as Figure 3 As shown, a neuron structure is shown, where x1, ..., x n They represent input signals respectively, n represents the number of input signals; w 1j ,…,w ij ,…,w nj They represent the weight corresponding to the first input signal, …, the weight corresponding to the i-th input signal, …, the weight corresponding to the n-th input signal; b j represents the bias value; f(.) represents the activation function used by the neuron. The output y of each neuron j Calculated by the following formula:

[0077]

[0078] In an embodiment of the present invention, historical credit data may be collected, including but not limited to: the customer's loan record, repayment behavior, financial status, credit history, etc. After the data is collected, the data is preprocessed to convert the data into a format that the model can understand and process to obtain training data. Data preprocessing includes operations such as data cleaning (removing outliers and missing values), feature encoding (converting categorical variables to numerical values), and data standardization (making numerical features of the same magnitude). Then, based on the obtained training data, the initial credit scoring model is evolved using a neuroevolution strategy to obtain the optimal target credit scoring model.

[0079] In this embodiment, by constructing an initial credit scoring model of a multi-layer neural network, preprocessing the historical credit data, and then optimizing the model with the help of a neural evolution strategy, a credit scoring model construction process that can automatically learn, adapt and optimize is realized. Not only can the structure and weight of the model be automatically optimized, but also manual intervention is reduced, costs are saved, and the efficiency of model optimization is improved.

[0080] In order to improve the accuracy of determining training data, in the neural evolution-based credit scoring method provided in Example 1 of the present application, the historical credit data is cleaned to obtain cleaned historical credit data; each historical credit record in the cleaned historical credit data is labeled to obtain a credit label corresponding to each historical credit record, wherein the credit label includes: a normal label or a risk label; based on the credit label corresponding to each historical credit record, the cleaned historical credit data is screened to obtain target historical credit data, wherein the number of historical credit records corresponding to the normal labels contained in the target historical credit data is equal to the number of historical credit records corresponding to the risk labels; the target historical credit data is standardized to obtain training data.

[0081] In the embodiment of the present invention, the collected historical credit data can be cleaned first to ensure the accuracy and completeness of the data and improve the efficiency and effect of model training. Data cleaning includes: removing duplicate records, processing missing values, identifying and correcting abnormal data points, etc., and also includes: removing noise in the data.

[0082] Then, each historical credit record in the cleaned historical credit data can be labeled to obtain the credit label corresponding to each historical credit record (divided into: normal label or risk label) to distinguish customers with good credit and customers with high credit risk. In this embodiment, the purpose of labeling is to assign a category label to each historical credit record to clearly distinguish the credit status and provide clear guidance for subsequent model training. The labeling process can mark each record as a "normal label" or "risk label" based on the repayment status of the loan, overdue history, credit limit usage, etc. Among them, the normal label: indicates that the customer has a good credit status, repays on time, and has no overdue records. The risk label: indicates that the customer has a credit risk and may have bad records such as overdue repayments and high debt ratios.

[0083] Afterwards, in order to ensure the balance of training data and avoid bias in the model when processing imbalanced data, the labeled data needs to be screened to ensure that the number of samples with normal labels and risk labels is equal, that is, based on the credit label corresponding to each historical credit record, the cleaned historical credit data can be screened to obtain the target historical credit data containing the same number of historical credit records corresponding to normal labels and the same number of historical credit records corresponding to risk labels. In this way, it can be ensured that the model can learn the characteristics of the two types of samples during training, avoid excessive bias towards one type of sample, and improve the generalization ability and fairness of the model.

[0084] In the embodiment of the present invention, the target historical credit data can be standardized, that is, the data is converted to a uniform scale to avoid the influence of the magnitude difference between features on the model training. Standardization processing includes: data scaling and data encoding (such as converting categorical data into numerical form). By scaling and encoding, the data is converted to a uniform scale, thereby ensuring that the model will not be adversely affected by the difference in feature magnitude when processing data, thereby improving the stability and accuracy of the model.

[0085] In this embodiment, the quality and balance of the training data used for training the credit scoring model are ensured through data cleaning, labeling, balanced screening and standardization, which is conducive to improving the reliability of the model and ensuring that it can make fair and accurate assessments when facing customers with different credit statuses.

[0086] In order to further accurately train the target credit scoring model, in the neural evolution-based credit scoring method provided in Example 1 of the present application, a preset number of individuals are determined to obtain an initialized population, wherein each individual represents an initial credit scoring model; each individual is encoded to obtain a neural network gene corresponding to the individual, wherein the neural network gene includes: multiple node genes and connection information corresponding to each node gene; each neural network gene is evolved to obtain an evolved neural network gene; based on the training data, a first fitness value of each evolved neural network gene is determined; an evolved neural network gene corresponding to a first fitness value greater than a preset fitness threshold is selected, and the selected evolved neural network gene is evolved until the number of evolutions reaches a preset number of evolutions, and the evolution is stopped to obtain a preset neural network gene; based on the training data, a second fitness value of each preset neural network gene is determined, and the preset neural network gene corresponding to the maximum second fitness value is determined as the target neural network gene; based on the target neural network gene, a target credit scoring model is determined.

[0087] In an embodiment of the present invention, at the beginning of the neural evolution, a preset number of individuals (i.e., a preset number), for example, 100 individuals, may be set, and these individuals constitute the initialization population. Each individual is essentially an initial credit scoring model, containing different neural network structures and weights. Then, each individual may be encoded to generate corresponding neural network genes (including: node genes and connection information). Among them, the node genes represent the neurons in the model, and the connection information describes the connection mode and weights between the nodes.

[0088] In the embodiment of the present invention, the neural network gene will be optimized through the evolution process of the genetic algorithm, including operations such as selection, crossover and mutation. Through the operation of the genetic algorithm, the neural network gene continues to evolve and can find a better model structure and parameters.

[0089] In an embodiment of the present invention, each evolved individual (i.e., evolved neural network gene) can be evaluated according to the training data set to obtain the first fitness value corresponding to each evolved neural network gene. The fitness value reflects the performance of the model on the data set and can be determined based on the prediction accuracy and other performance indicators of the model. Then, the individuals whose first fitness value exceeds the preset fitness threshold (which can be adjusted according to actual conditions) are screened out, and these individuals will enter the next round of evolution. After that, the screened individuals continue to evolve until the number of evolutions reaches the preset number of evolutions (i.e., the number of iterations of the evolutionary process is controlled to avoid infinite iterations, while ensuring that the model has enough time to optimize). In this process, the individuals in the population will be continuously optimized to improve their fitness values ​​on the training data.

[0090] In the embodiment of the present invention, after the preset evolution times are completed, the second fitness value of each preset neural network gene after evolution can be evaluated, and the individual with the highest second fitness value is selected, and its neural network gene is determined as the target neural network gene. Finally, the target neural network gene is converted into an actual credit scoring model for evaluating the credit status of the customer.

[0091] In this embodiment, the credit scoring model is optimized through the neural evolution strategy, starting from the initialization of the population, going through the steps of encoding, evolution, and evaluation, and finally selecting the optimal neural network gene to construct the target credit scoring model. In this way, the accuracy and generalization ability of the model are effectively improved, and the dependence on manual experience is reduced. At the same time, by setting the adaptation threshold and the number of evolutions, the controllability and efficiency of the evolution process are guaranteed, and overfitting and invalid iterations are avoided.

[0092] Figure 4 is a schematic diagram of a neural network encoding according to Embodiment 1 of the present invention, such as Figure 4 As shown in the figure, a neural network can have two forms: genotype and phenotype, where the genotype includes: node genes and corresponding connection genes; the phenotype is visualized through nodes and lines. For example, the node genes of the genotype of a neural network are divided into: No. 1, type: input; No. 2, type: input; No. 3, type: input; No. 3, type: hidden; No. 5, type: output; the connection genes are divided into: input: 1, output: 4, weight: 0.7, available: yes, innovation number (i.e. used to track the source of a connection): 1; input: 2, output: 4, weight: 0.3, available: yes, innovation number: 2; input: 1, output: 4, weight: 0.7, available: no, innovation number: 3; input: 3, output: 5, weight: 0.2, available: yes, innovation number: 4; input: 4, output: 5, weight: 0.6, available: yes, innovation number: 5. The corresponding phenotype of this neural network is: node 1 and node 2 are connected to node 4, node 4 is connected to node 5, and node 3 is connected to node 5.

[0093] In order to improve the accuracy of determining the neural network genes corresponding to an individual, in the neural evolution-based credit scoring method provided in Example 1 of the present application, the coordinate parameters of each neuron in the initial credit scoring model represented by the individual are determined, wherein the coordinate parameters include: layer parameters and bit parameters; all neurons are combined in pairs to obtain multiple pairs of neuron pairs; the coordinate parameters of the two neurons corresponding to each pair of neurons are input into the composite pattern generation network to obtain the weight between the two neurons; each neuron is characterized as a node gene, and the weight between the neuron and another neuron is characterized as connection information; based on all node genes and the connection information corresponding to each node gene, the neural network gene is determined.

[0094] In an embodiment of the present invention, the position of each neuron in the model in the neural network, i.e., the coordinate parameters, can be set first. These parameters include layer parameters (indicating the layer where the neuron is located, such as the input layer, hidden layer, or output layer) and position parameters (indicating the position in the same layer) to facilitate the subsequent calculation of weights through a composite pattern generating network (such as Compos it ional Pattern Producing Networks, CPPNs for short). Then the neurons in the model can be paired in pairs to form multiple neuron pairs. Afterwards, for each pair of neurons, their coordinate parameters can be input into a pre-trained composite pattern generating network (CPPNs), and the CPPNs will generate connection weights based on the input coordinate information.

[0095] In the embodiment of the present invention, each neuron can be regarded as a node gene, and the weights between neuron pairs are used as connection information. Then all node genes and their connection information are integrated to form a complete neural network gene. This gene contains the complete structural information of the model, that is, the positions of neurons and the connection weights between them.

[0096] In this embodiment, by using a composite pattern generation network to indirectly encode the connection weights of the neural network, the number of genes required for direct encoding can be reduced, thereby improving the efficiency of neural evolution.

[0097] In order to improve the accuracy of determining the evolved neural network genes, in the neural evolution-based credit scoring method provided in Example 1 of the present application, cross-evolution is performed on every two neural network genes according to a first preset probability; after cross-evolution is performed on all neural network genes, mutation evolution is performed on each neural network gene according to a second preset probability to obtain an evolved neural network gene.

[0098] In an embodiment of the present invention, in a population of neural evolution, two selected neural network genes can be cross-evolved according to a first preset probability, that is, the gene chains of the two selected individuals are cross-evolved according to a certain probability (cross-over probability Pc), thereby generating two new individuals. The cross-over position is random, and Pc is a preset system parameter. Only one cross-over point is randomly set in the individual coding string, at which part of the chromosomes of the two ligand individuals are exchanged with each other. Then, each neural network gene can be mutated, wherein the mutation will be performed at a second preset probability, which means that each gene has a specific probability of being randomly changed, that is, the gene values ​​on certain loci of the individual strings in the population are changed. In the selected individuals, each bit of the gene chain of the new individual is transformed in an opposite direction according to the mutation probability (Pm). The mutation operation is also based on probability, and Pm is generally set very small (usually Pm≤0.05). The value of a certain bit or several bits of the gene loci randomly specified by the mutation probability in the individual coding string is subjected to mutation operation.

[0099] In the embodiment of the present invention, new neural network genes can be generated through the above-mentioned crossover and mutation operations, and these genes form the basis of the next generation population.

[0100] In this embodiment, efficient optimization of neural network genes is achieved through crossover and mutation evolution strategies. Crossover evolution is carried out in a probabilistic form, and new individuals can be created through gene recombination, which may combine the advantages of different individuals. Mutation evolution is also carried out in a probabilistic form, and new features are introduced by randomly changing genes to increase population diversity.

[0101] In order to improve the accuracy of determining the first fitness value of the evolved neural network gene, in the neural evolution-based credit scoring method provided in Example 1 of the present application, each evolved neural network gene is converted into an evolved credit scoring model; the evolved credit scoring model is used to process each historical credit record in the training data to obtain a predicted scoring result corresponding to the historical credit record, wherein the predicted scoring result includes: a normal credit record or a risky credit record; based on the predicted scoring result corresponding to each historical credit record and the pre-marked credit label, the index value of the evolved credit scoring model is determined; and the index value is determined as the first fitness value of the evolved neural network gene.

[0102] In the embodiment of the present invention, each evolved neural network gene can be converted into a specific evolved credit scoring model, that is, the adjusted neural network gene will be decoded into a credit scoring model containing a specific structure and weights. Then, the evolved credit scoring model is used to process each historical credit record in the training data to obtain the predicted scoring result corresponding to the historical credit record. The predicted scoring result is the model's prediction of the historical credit record, which is used to determine whether the credit record is a normal credit record or a risky credit record.

[0103] In an embodiment of the present invention, the predicted scoring results of the model can be compared with the pre-labeled credit labels of each historical credit record in the training data to calculate the performance indicators (i.e., indicator values) of the model, such as accuracy, recall rate, F1 score, etc. These indicators will be used to quantify the prediction performance of the model. Then, the calculated performance indicator value is used as the first fitness value of the evolved neural network gene to evaluate the performance of the individual in the credit scoring task. This fitness value will become the standard for selecting and eliminating individuals in the genetic algorithm to ensure that the population evolves in an optimized direction.

[0104] For example, the accuracy rate can be used as an indicator for evaluating the model, and the accuracy rate calculation formula is as follows:

[0105]

[0106] Among them, TP (True Positive) represents the number of correctly classified positive samples (i.e., the number of normal historical credit records whose predicted scoring results are normal and whose credit labels are also normal), TN (True Negative) represents the number of correctly classified negative samples (i.e., the number of risky historical credit records whose predicted scoring results are risky and whose credit labels are also risky), FP (Failed Positive) represents the number of incorrectly classified positive samples (i.e., the number of normal historical credit records whose predicted scoring results are risky but whose credit labels are normal), and FN (Failed Negative) represents the number of incorrectly classified negative samples (i.e., the number of risky historical credit records whose predicted scoring results are normal but whose credit labels are risky).

[0107] The neuroevolution-based credit scoring method provided in the embodiment of the present application can evolve the neural network through the neuroevolution strategy to obtain a target credit scoring model, and then process the credit data of the target object through the target credit scoring model to obtain the scoring result of the target object, thereby achieving the technical effect of efficiently and accurately constructing a credit scoring model, and accurately and quickly evaluating the credit score of a customer through the credit scoring model, thereby solving the technical problem of low accuracy in credit scoring of customers in related technologies.

[0108] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0109] Example 2

[0110] The embodiment of the present application also provides a credit scoring device based on neuroevolution. It should be noted that the credit scoring device based on neuroevolution in the embodiment of the present application can be used to execute the credit scoring method based on neuroevolution provided in the embodiment of the present application. The credit scoring device based on neuroevolution provided in the embodiment of the present application is introduced below.

[0111] According to an embodiment of the present application, a device for implementing the above-mentioned neural evolution-based credit scoring method is also provided. Figure 5 is a schematic diagram of an optional neural evolution-based credit scoring device according to an embodiment of the present invention, such as Figure 5 As shown, the credit scoring device may include: an acquisition unit 50, a processing unit 51, and an input unit 52.

[0112] Wherein, the acquisition unit 50 is used to acquire a target credit scoring model constructed based on the neuroevolution strategy;

[0113] The processing unit 51 is used to obtain the credit data of the target object and process the credit data to obtain the target credit data;

[0114] The input unit 52 is used to input the target credit data into the target credit scoring model and output the scoring result for the target object.

[0115] The neuroevolution-based credit scoring device provided in the embodiment of the present application can evolve the neural network through the neuroevolution strategy to obtain a target credit scoring model, and then process the credit data of the target object through the target credit scoring model to obtain the scoring result of the target object, thereby achieving the technical effect of efficiently and accurately constructing a credit scoring model, and accurately and quickly evaluating the credit score of a customer through the credit scoring model, thereby solving the technical problem of low accuracy in credit scoring of customers in related technologies.

[0116] Optionally, the credit scoring device also includes: a construction unit, which is used to construct an initial credit scoring model before obtaining a target credit scoring model constructed based on a neural evolution strategy, wherein the model structure of the initial credit scoring model includes at least: a multi-layer neural network layer, the neural network layer includes at least: an input layer, a fully connected layer, and an output layer, each neural network layer includes: a plurality of neurons, and there are corresponding weights between neurons in each neural network layer and neurons in an adjacent neural network layer, and the adjacent neural network layer refers to the previous neural network layer or the next neural network layer of the neural network layer; a collection unit, which is used to collect historical credit data and process the historical credit data to obtain training data; an evolution unit, which is used to evolve the initial credit scoring model based on the training data using a neural evolution strategy to obtain a target credit scoring model.

[0117] Optionally, the collection unit includes: a first cleaning module, used to clean the historical credit data to obtain the cleaned historical credit data; a first labeling module, used to label each historical credit record in the cleaned historical credit data to obtain a credit label corresponding to each historical credit record, wherein the credit label includes: a normal label or a risk label; a first screening module, used to screen the cleaned historical credit data based on the credit label corresponding to each historical credit record to obtain target historical credit data, wherein the number of historical credit records corresponding to the normal label contained in the target historical credit data is equal to the number of historical credit records corresponding to the risk label; a first processing module, used to perform standardization processing on the target historical credit data to obtain training data.

[0118] Optionally, the evolution unit includes: a first determination module, used to determine a preset number of individuals to obtain an initialized population, wherein each individual represents an initial credit scoring model; a first encoding module, used to encode each individual to obtain a neural network gene corresponding to the individual, wherein the neural network gene includes: multiple node genes and connection information corresponding to each node gene; a first evolution module, used to evolve each neural network gene to obtain an evolved neural network gene; a second determination module, used to determine the first fitness value of each evolved neural network gene based on training data; a first selection module, used to select an evolved neural network gene corresponding to a first fitness value greater than a preset fitness threshold, and evolve the selected evolved neural network gene until the number of evolutions reaches a preset number of evolutions, stops the evolution, and obtains the preset neural network gene; a third determination module, used to determine the second fitness value of each preset neural network gene based on the training data, and determine the preset neural network gene corresponding to the maximum second fitness value as the target neural network gene; a fourth determination module, used to determine the target credit scoring model based on the target neural network gene.

[0119] Optionally, the first encoding module includes: a first determination submodule, used to determine the coordinate parameters of each neuron in the initial credit scoring model of the individual representative, wherein the coordinate parameters include: layer parameters and bit parameters; a first combination submodule, used to combine all neurons in pairs to obtain multiple pairs of neuron pairs; a first input submodule, used to input the coordinate parameters of the two neurons corresponding to each pair of neurons into the composite pattern generation network to obtain the weight between the two neurons; a first characterization submodule, used to characterize each neuron as a node gene, and characterize the weight between the neuron and another neuron as connection information; a second determination submodule, used to determine the neural network gene based on all node genes and the connection information corresponding to each node gene.

[0120] Optionally, the first evolution module includes: a first evolution submodule, used to perform cross-evolution on every two neural network genes according to a first preset probability; and a second evolution submodule, used to perform mutation evolution on each neural network gene according to a second preset probability after cross-evolution on all neural network genes, to obtain evolved neural network genes.

[0121] Optionally, the second determination module includes: a first conversion submodule, used to convert each evolved neural network gene into an evolved credit scoring model; a first processing submodule, used to use the evolved credit scoring model to process each historical credit record in the training data to obtain a predicted scoring result corresponding to the historical credit record, wherein the predicted scoring result includes: a normal credit record or a risky credit record; a third determination submodule, used to determine an index value of the evolved credit scoring model based on the predicted scoring result corresponding to each historical credit record and a pre-labeled credit label; and a fourth determination submodule, used to determine the index value as the first fitness value of the evolved neural network gene.

[0122] The credit scoring device may further include a processor and a memory. The acquisition unit 50, the processing unit 51, the input unit 52, etc. are all stored in the memory as program units, and the processor executes the program units stored in the memory to implement corresponding functions.

[0123] The processor includes a kernel, which retrieves the corresponding program unit from the memory. One or more kernels can be set, and the target credit data is input into the target credit scoring model by adjusting the kernel parameters, and the scoring result of the target object is output.

[0124] The above-mentioned memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash memory (fl ash RAM), and the memory includes at least one storage chip.

[0125] It should be noted that the acquisition unit 50, the processing unit 51, and the input unit 52 correspond to steps S201 to S203 in Example 1, and the examples and application scenarios implemented by the above units and the corresponding steps are the same, but are not limited to the contents disclosed in the above Example 1. It should be noted that the above units can be hardware components or software components stored in a memory (e.g., memory 104) and processed by one or more processors (e.g., processors 102a, 102b, ..., 102n), and the above units can also be run in the computer terminal 10 provided in Example 1 as part of the device.

[0126] Example 3

[0127] An embodiment of the present application may provide an electronic device, Figure 6 is a structural block diagram of an electronic device according to an embodiment of the present invention. Figure 6 As shown, the electronic device may include: one or more ( Figure 6 (only one is shown) processor 602, memory 604, storage controller, and peripheral interface, wherein the peripheral interface is connected to the radio frequency module, audio module and display.

[0128] Among them, the memory can be used to store software programs and modules, such as the program instructions / modules corresponding to the credit scoring method and device based on neuroevolution in the embodiments of the present application. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory, that is, realizing the above-mentioned credit scoring method based on neuroevolution. The memory may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory may further include a memory remotely arranged relative to the processor, and these remote memories can be connected to the terminal via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0129] The processor can call the information and application programs stored in the memory through the transmission device to perform the following steps: obtain a target credit scoring model built based on the neuroevolution strategy; obtain the credit data of the target object and process the credit data to obtain the target credit data; input the target credit data into the target credit scoring model and output the scoring result for the target object.

[0130] The processor can also call the information and application programs stored in the memory through the transmission device to perform the following steps: construct an initial credit scoring model, wherein the model structure of the initial credit scoring model includes at least: a multi-layer neural network layer, the neural network layer includes at least: an input layer, a fully connected layer, and an output layer, each neural network layer includes: a plurality of neurons, and there are corresponding weights between the neurons in each neural network layer and the neurons in the adjacent neural network layer, and the adjacent neural network layer refers to the previous neural network layer or the next neural network layer of the neural network layer; collect historical credit data, and process the historical credit data to obtain training data; based on the training data, use a neural evolution strategy to evolve the initial credit scoring model to obtain a target credit scoring model.

[0131] The processor can also call the information and application programs stored in the memory through the transmission device to perform the following steps: clean the historical credit data to obtain cleaned historical credit data; mark each historical credit record in the cleaned historical credit data to obtain a credit label corresponding to each historical credit record, wherein the credit label includes: a normal label or a risk label; based on the credit label corresponding to each historical credit record, filter the cleaned historical credit data to obtain target historical credit data, wherein the number of historical credit records corresponding to the normal labels contained in the target historical credit data is equal to the number of historical credit records corresponding to the risk labels; standardize the target historical credit data to obtain training data.

[0132] The processor can also call the information and application stored in the memory through the transmission device to perform the following steps: determine a preset number of individuals to obtain an initialized population, wherein each individual represents an initial credit scoring model; encode each individual to obtain a neural network gene corresponding to the individual, wherein the neural network gene includes: multiple node genes and connection information corresponding to each node gene; evolve each neural network gene to obtain an evolved neural network gene; based on the training data, determine the first fitness value of each evolved neural network gene; select the evolved neural network gene corresponding to the first fitness value greater than the preset fitness threshold, and evolve the selected evolved neural network gene until the number of evolutions reaches the preset number of evolutions, stop the evolution, and obtain the preset neural network gene; based on the training data, determine the second fitness value of each preset neural network gene, and determine the preset neural network gene corresponding to the maximum second fitness value as the target neural network gene; based on the target neural network gene, determine the target credit scoring model.

[0133] The processor can also call the information and application programs stored in the memory through the transmission device to perform the following steps: determine the coordinate parameters of each neuron in the initial credit scoring model represented by the individual, wherein the coordinate parameters include: layer parameters and bit parameters; combine all neurons in pairs to obtain multiple pairs of neurons; input the coordinate parameters of the two neurons corresponding to each pair of neurons into the composite pattern generation network to obtain the weight between the two neurons; characterize each neuron as a node gene, and characterize the weight between the neuron and another neuron as connection information; determine the neural network gene based on all node genes and the connection information corresponding to each node gene.

[0134] The processor can also call the information and application programs stored in the memory through the transmission device to execute the following steps: cross-evolve every two neural network genes according to a first preset probability; after cross-evolving all neural network genes, mutate and evolve each neural network gene according to a second preset probability to obtain an evolved neural network gene.

[0135] The processor can also call the information and application programs stored in the memory through the transmission device to perform the following steps: convert each evolved neural network gene into an evolved credit scoring model; use the evolved credit scoring model to process each historical credit record in the training data to obtain a predicted scoring result corresponding to the historical credit record, wherein the predicted scoring result includes: a normal credit record or a risky credit record; determine the index value of the evolved credit scoring model based on the predicted scoring result corresponding to each historical credit record and the pre-marked credit label; determine the index value as the first fitness value of the evolved neural network gene.

[0136] By adopting the embodiment of the present application, a credit scoring scheme based on neuroevolution is provided. By evolving the neural network through the neuroevolution strategy, a target credit scoring model can be obtained, and then the credit data of the target object is processed through the target credit scoring model to obtain the scoring result of the target object, thereby achieving the technical effect of efficiently and accurately constructing a credit scoring model, and accurately and quickly evaluating the credit score of a customer through the credit scoring model, thereby solving the technical problem of low accuracy in credit scoring of customers in related technologies.

[0137] It can be understood by those skilled in the art that Figure 6 The structure shown is for illustration only, and the electronic device may also be a terminal device such as a smart phone, a tablet computer, a PDA, and a Mobile Internet Device (MID). Figure 6 The structure of the electronic device is not limited. Figure 6 More or fewer components (such as network interfaces, display devices, etc.) shown in, or having Figure 6 Different configurations are shown.

[0138] A person of ordinary skill in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing the hardware related to the terminal device through a program, and the program can be stored in a computer-readable storage medium, and the storage medium may include: a flash drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0139] Example 4

[0140] The embodiment of the present application further provides a storage medium. Optionally, in this embodiment, the storage medium can be used to store the program code executed by the credit scoring method based on neuroevolution provided in the first embodiment.

[0141] Optionally, in this embodiment, the above storage medium may be located in any computer terminal in a computer terminal group in a computer network, or in any mobile terminal in a mobile terminal group.

[0142] The present application also provides a computer program product, which, when executed on a data processing device, is suitable for executing the program steps of the neuroevolution-based credit scoring method.

[0143] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0144] In the above embodiments of the present application, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0145] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0146] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0147] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0148] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, server or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, disk or CD-ROM and other media that can store program codes.

[0149] The above is only a preferred implementation of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A credit scoring method based on neuroevolution, characterized in that: include: Obtain a target credit scoring model built based on a neuroevolution strategy; Acquiring credit data of a target object, and processing the credit data to obtain target credit data; The target credit data is input into the target credit scoring model, and a scoring result for the target object is output.

2. The credit scoring method according to claim 1, characterized in that: Before obtaining the target credit scoring model built based on the neuroevolution strategy, it also includes: Constructing an initial credit scoring model, wherein the model structure of the initial credit scoring model at least includes: a multi-layer neural network layer, the neural network layer at least includes: an input layer, a fully connected layer, and an output layer, each neural network layer includes: a plurality of neurons, and the neurons in each neural network layer have corresponding weights with the neurons in an adjacent neural network layer, and the adjacent neural network layer refers to the neural network layer of the previous neural network layer or the neural network layer of the next neural network layer; Collecting historical credit data and processing the historical credit data to obtain training data; Based on the training data, the initial credit scoring model is evolved using the neuroevolution strategy to obtain the target credit scoring model.

3. The credit scoring method according to claim 2, characterized in that: The step of processing the historical credit data to obtain training data includes: Cleaning the historical credit data to obtain cleaned historical credit data; Marking each historical credit record in the cleaned historical credit data to obtain a credit label corresponding to each historical credit record, wherein the credit label includes: a normal label or a risk label; Based on the credit label corresponding to each of the historical credit records, the cleaned historical credit data is screened to obtain target historical credit data, wherein the number of the historical credit records corresponding to the normal label contained in the target historical credit data is equal to the number of the historical credit records corresponding to the risk label; The target historical credit data is standardized to obtain training data.

4. The credit scoring method according to claim 2, characterized in that: The step of evolving the initial credit scoring model using the neural evolution strategy based on the training data to obtain the target credit scoring model includes: Determining a preset number of individuals to obtain an initial population, wherein each individual represents an initial credit scoring model; Encoding each of the individuals to obtain a neural network gene corresponding to the individual, wherein the neural network gene includes: a plurality of node genes and connection information corresponding to each of the node genes; Evolving each of the neural network genes to obtain evolved neural network genes; Based on the training data, determining a first fitness value of each of the evolved neural network genes; Selecting the evolved neural network gene corresponding to the first fitness value greater than a preset fitness threshold, and evolving the selected evolved neural network gene until the number of evolutions reaches a preset number of evolutions, stopping the evolution, and obtaining a preset neural network gene; Based on the training data, determining the second fitness value of each of the preset neural network genes, and determining the preset neural network gene corresponding to the maximum second fitness value as the target neural network gene; Based on the target network gene, the target credit score model is determined.

5. The credit scoring method according to claim 4, characterized in that: The step of encoding each individual to obtain a neural network gene corresponding to the individual comprises: Determining coordinate parameters of each neuron in the initial credit scoring model represented by the individual, wherein the coordinate parameters include: layer parameters and bit parameters; Combining all the neurons in pairs to obtain multiple pairs of neurons; Inputting the coordinate parameters of the two neurons corresponding to each pair of the neurons into a composite pattern generation network to obtain a weight between the two neurons; Characterize each of the neurons as the node gene, and characterize the weight between the neuron and another neuron as the connection information; The neural network gene is determined based on all the node genes and the connection information corresponding to each of the node genes.

6. The credit scoring method according to claim 4, characterized in that: The step of evolving each of the neural network genes to obtain evolved neural network genes comprises: Performing crossover evolution on every two neural network genes according to a first preset probability; After cross evolution is performed on all the neural network genes, each of the neural network genes is subjected to mutation evolution according to a second preset probability to obtain the evolved neural network gene.

7. The credit scoring method according to claim 4, characterized in that: The step of determining a first fitness value of each of the evolved neural network genes based on the training data comprises: converting each of the evolved neural network genes into an evolved credit scoring model; Using the evolutionary credit scoring model to process each historical credit record in the training data, obtaining a predicted scoring result corresponding to the historical credit record, wherein the predicted scoring result includes: a normal credit record or a risky credit record; Determining the index value of the evolutionary credit scoring model based on the predicted scoring result corresponding to each of the historical credit records and the pre-annotated credit label; The index value is determined as the first fitness value of the evolved neural network gene.

8. A credit scoring device based on neural evolution, characterized in that: include: An acquisition unit, used for acquiring a target credit scoring model built based on a neuroevolution strategy; A processing unit, used to obtain credit data of a target object and process the credit data to obtain target credit data; The input unit is used to input the target credit data into the target credit scoring model and output the scoring result of the target object.

9. A computer program product, characterized in that It comprises a non-volatile computer-readable storage medium, wherein the non-volatile computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the neuroevolution-based credit scoring method according to any one of claims 1 to 7 is implemented.

10. An electronic device, characterized in that: It includes one or more processors and a memory, wherein the memory is used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the neuroevolution-based credit scoring method as described in any one of claims 1 to 7.

Images