Enterprise credit evaluation system and method

Abstract

The invention relates to an enterprise credit evaluation system and method, the system comprises a data acquisition module, a feature engineering module and an evaluation module which are connected in sequence, and the method comprises the steps: obtaining the local data of an enterprise and the data of a regional big data center, and converting the data of the regional big data center into the data of an enterprise big data center; carrying out entropy calculation, IV value calculation and correlation coefficient analysis on the obtained data to determine in-mold multi-dimensional information, carrying out feature screening, screening out important features from the in-mold multi-dimensional information, and constructing a feature set; training an XGBoost model by using the feature set to obtain an enterprise credit evaluation model; and inputting the in-mold multi-dimensional information and the unified social credit code information of the to-be-assessed enterprise into the enterprise credit assessment model, and outputting a credit score and a suggested credit line corresponding to the to-be-assessed enterprise. Compared with the prior art, the method has the advantages that the problems of dispersion and inconsistency of multi-channel data sources in the prior art can be solved, timely updating of the data is ensured, and the accuracy of enterprise credit assessment results is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of big data processing, and in particular to an enterprise credit evaluation system and method. Background Art

[0002] Enterprise credit score is a classification of enterprise credit according to the results of enterprise credit assessment, which reflects the level of enterprise credit. Different credit ratings will greatly vary the difficulty and conditions for enterprises to raise funds in the market and obtain bank loans. The assessment of enterprise credit rating mainly includes tax registration, tax declaration, account books, voucher management, tax payment, etc.

[0003] In the early days of corporate credit assessment, more relied on the subjective judgment of loan officers, combining information such as the company's credit history, financial data and collateral to assess risk. However, with the advancement of big data and artificial intelligence technology, corporate credit assessment is increasingly moving towards a data-driven decision-making model. This shift has greatly improved the efficiency of the loan approval process and also enhanced the accuracy of risk control.

[0004] Existing research uses data models to score corporate credit. For example, the corporate loan scoring model is a quantitative tool that aims to evaluate a company's financial status, operating risks, and the impact of the external environment on its repayment ability. The model analyzes data on the company's financial statements, credit records, industry conditions, and management levels to provide financial institutions with a quantitative scoring result. This score is usually used as a key basis for financial institutions to determine whether to lend to the company and determine the loan conditions.

[0005] However, in actual applications, the data quality and standardization of financial statements among enterprises are often uneven, especially in small and medium-sized enterprises. Many small and medium-sized enterprises lack a sound financial management system due to their small scale, short operation time, and lack of a sound financial management system, resulting in different standards for the preparation of their financial statements, and sometimes even unable to provide complete and accurate financial data. Some enterprises may not have a dedicated financial team, and the recording and collation of financial data may rely on external accounting firms or temporary personnel, which may easily lead to incomplete data or untimely updates. In addition, due to the funding and resource constraints faced by small and medium-sized enterprises, they usually lack a sound internal audit mechanism and cannot effectively review and standardize their financial reports. This situation poses a challenge to the operation of loan application scoring models, because models often rely on accurate and standardized financial data to assess the credit risk of enterprises. If the data provided by the enterprise is incomplete or inaccurate, the prediction results of the model may lose their reference value, thereby increasing the loan risk of financial institutions. Summary of the invention

[0006] The purpose of the present invention is to provide a corporate credit assessment system and method in order to overcome the defects of the above-mentioned prior art, which can solve the problems of dispersion and inconsistency of multi-channel data sources in the prior art, ensure timely data updating, and improve the accuracy of corporate credit assessment results.

[0007] The object of the present invention can be achieved by the following technical solutions: an enterprise credit evaluation system, comprising a data acquisition module, a feature engineering module and an evaluation module connected in sequence, wherein the data acquisition module is used to update and obtain enterprise local data and regional big data center data, and convert the regional big data center data into enterprise big data center data;

[0008] The feature engineering module is used to mine input multi-dimensional information from enterprise local data and enterprise big data center data;

[0009] The evaluation module is used to filter out important features from the input multi-dimensional information, and to predict and output the credit score and recommended credit limit of the enterprise.

[0010] Furthermore, the data acquisition module is connected to the external data interface and the big data center respectively to obtain the enterprise local data and the regional big data center data accordingly.

[0011] Furthermore, the data acquisition module is provided with an enterprise regional database and a conversion unit, the enterprise regional database is used to store the regional related information of each enterprise, and the conversion unit calls the regional related information of each enterprise from the enterprise regional database to convert the regional big data center data into the enterprise big data center data.

[0012] A method for evaluating corporate credit, comprising the following steps:

[0013] S1. Obtain enterprise local data and regional big data center data respectively, and convert the regional big data center data into enterprise big data center data;

[0014] S2. Calculate the entropy value, IV (Information Value) value and analyze the correlation coefficient for the enterprise local data and enterprise big data to determine the multi-dimensional information to be input into the model;

[0015] S3, perform feature screening on the multi-dimensional information input into the model, select important features from it, and construct a feature set;

[0016] S4. Use the feature set to train the XGBoost model to obtain the enterprise credit assessment model;

[0017] S5. Input the multi-dimensional information of the enterprise to be evaluated and the unified social credit code information into the enterprise credit evaluation model, and output the credit score and recommended credit limit corresponding to the enterprise to be evaluated.

[0018] Furthermore, the step S1 specifically includes the following steps:

[0019] S11. Obtain enterprise local data through external data collection channels;

[0020] S12, constructing an enterprise regional database for storing regional related information of the enterprise;

[0021] S13. Obtain the financial statement information of the enterprise from the big data center, and associate it with the regional related information of the enterprise to convert it into the enterprise big data center data accordingly.

[0022] Furthermore, the enterprise local data includes the enterprise's industrial and commercial information, judicial information and patent information.

[0023] Furthermore, the financial statement information of the enterprise includes the enterprise's tax information, invoice information, financial information, social security data and provident fund data.

[0024] Furthermore, the specific process of step S12 includes:

[0025] First, through the address data interface, extract and call the detailed address information of each enterprise;

[0026] Then, the enterprise address information is dotted on the map, the actual location of each enterprise is marked on the map, and detailed information related to the area is extracted, including the name of the building, the park where it is located, the building area, the floor structure, the use function and the surrounding supporting facilities;

[0027] For enterprises in the park, collect general information about the entire park, including the area of ​​the park, the number of settled enterprises, industrial distribution and the operation and management information of the park;

[0028] Finally, all data are systematically organized and stored to construct an enterprise regional database, which is used to store the buildings where each enterprise is located and their geographical location, as well as the density of enterprises in the region, the types of buildings, and the industry distribution information of the park.

[0029] Furthermore, the specific process of step S2 includes:

[0030] Entropy values ​​are calculated for enterprise local data and enterprise big data, wherein the larger the entropy value, the more dispersed and uncertain the information provided by the feature; the smaller the entropy value, the higher the information concentration of the feature, and the more certain the information provided; data information with an entropy value less than a preset first threshold is retained;

[0031] Calculate the IV value for the enterprise local data and enterprise big data, and retain the data information whose IV value is higher than the preset second threshold;

[0032] Conduct correlation coefficient analysis on enterprise local data and enterprise big data, and delete redundant data information with multicollinearity.

[0033] Furthermore, the step S3 specifically uses the random forest algorithm to perform feature screening on the input multi-dimensional information to screen out the important features that have the greatest impact on the corporate credit score.

[0034] Compared with the prior art, the present invention has the following advantages:

[0035] The present invention sets a data acquisition module, a feature engineering module and an evaluation module connected in sequence, uses the data acquisition module to update and obtain the enterprise local data and the regional big data center data, and converts the regional big data center data into the enterprise big data center data; uses the feature engineering module to mine the input model multi-dimensional information from the enterprise local data and the enterprise big data center data; uses the evaluation module to screen out important features from the input model multi-dimensional information, and predicts and outputs the credit score and recommended credit limit of the enterprise. In this way, the multi-dimensional massive data information of the enterprise can be comprehensively obtained, and valuable input model variables can be mined from it, and then the evaluation model is constructed based on it, thereby improving the accuracy of the enterprise credit evaluation results.

[0036] The present invention constructs an enterprise regional database to store the regional related information of the enterprise, obtains the financial statement information of the enterprise from the big data center, and then converts it according to the enterprise building relationship of the enterprise regional database, so as to generate big data center data of the enterprise dimension, which is beneficial to the accuracy of subsequent modeling.

[0037] The present invention obtains the financial statements of enterprises through the big data center channel, which can greatly improve the quality, integrity and real-time nature of the data. It not only solves the problems of dispersion and inconsistency of multi-channel data sources in the prior art, but also ensures timely updating of data and improves the prediction accuracy of the model. By automatically obtaining the financial statements of small and medium-sized enterprises, it can reduce manual intervention and improve efficiency.

[0038] The present invention performs entropy calculation, IV value calculation and correlation coefficient analysis on enterprise local data and enterprise big data to determine the multi-dimensional information input into the model. The entropy value calculation is used to measure the information content and uncertainty of each feature; the IV value calculation is used to effectively reduce irrelevant or weakly correlated features, thereby avoiding model complexity and overfitting problems; and the correlation coefficient analysis is used to identify which variables have multicollinearity problems, avoid feature redundancy and multicollinearity, and improve the accuracy and robustness of the model.

[0039] After feature engineering is completed, the present invention continues to perform feature screening on the multi-dimensional information input into the model, screens out important features, and uses the feature importance calculation of the random forest to screen out the features that have the greatest influence on the enterprise score, thereby enhancing the interpretability of the model; since the random forest uses the average results of multiple decision trees, it can well avoid the overfitting problem of a single decision tree and improve the generalization ability of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 It is a schematic diagram of the method flow of the present invention;

[0041] Figure 2 Schematic diagram of the application framework of the embodiment. DETAILED DESCRIPTION

[0042] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0043] Example

[0044] In the enterprise loan scoring model, multi-dimensional information is usually covered, mainly including the following aspects:

[0045] (1) Financial factors

[0046] Financial factors are one of the most core parts of the corporate loan scoring model because they directly reflect the company's debt repayment ability and financial health. The main financial indicators include:

[0047] 1. Liquidity indicators: such as current ratio and quick ratio, measure the company's short-term debt repayment ability. If the company's current assets are insufficient, it may be difficult to cope with short-term debt pressure.

[0048] 2. Profitability indicators: such as gross profit margin, net profit margin and earnings before interest and taxes (EBITDA), these indicators reflect the operating efficiency and profitability of the company.

[0049] 3. Leverage ratio: Such as debt-to-asset ratio and equity multiplier, assess the debt level of the enterprise. Too high a leverage ratio may indicate that the enterprise relies on borrowing to operate, increasing the risk of default.

[0050] 4. Cash flow: The cash flow status of an enterprise, especially operating cash flow, directly affects its debt repayment ability.

[0051] (2) Credit history

[0052] Credit history is one of the important indicators in the corporate loan scoring model, which reflects the company's past credit performance. Banks usually obtain the company's credit record through credit reports, including its loan repayment status, historical default or delinquency records. Companies with good credit histories are usually considered to have lower risks, while companies with multiple default records may be considered high-risk customers, facing higher loan costs or loan rejection.

[0053] (3) Industry and market factors

[0054] The industry and market environment in which the enterprise is located also have an important impact on loan approval. Some models will consider factors such as the industry life cycle, market competition, and industrial policy. For example, a high-tech industry enterprise in a growth period may receive a higher rating in the scoring model, while a traditional manufacturing industry enterprise in a recession period may receive a lower rating due to future uncertainties. In addition, external factors such as market demand, price fluctuations, and raw material supply will also affect the operating risk of the enterprise.

[0055] With the rapid development of big data technology, modern corporate loan application scoring models are increasingly relying on data analysis and artificial intelligence technologies. These technologies can improve the prediction accuracy of the model and help banks better assess risks by analyzing massive amounts of data. The following are some commonly used technical supports:

[0056] (1) Big Data Analysis

[0057] Big data technology enables financial institutions to collect and analyze a large amount of external data, which includes not only traditional financial statements, but also non-financial data, such as market dynamics, corporate online reputation, customer reviews, etc. By integrating this information, the scoring model can more comprehensively reflect the real operating conditions of the company.

[0058] (2) Machine Learning and Artificial Intelligence

[0059] Machine learning and artificial intelligence technologies play an increasingly important role in modern scoring models. By training on historical data, machine learning algorithms can discover hidden risk patterns and automatically update scoring criteria. This dynamic evaluation method enables the model to adapt to the ever-changing market environment and improve the accuracy of risk prediction.

[0060] (3) Cloud computing and data processing

[0061] The application of cloud computing makes the storage and processing of large-scale data more efficient. Financial institutions can quickly access and process data from different channels through cloud platforms and update scoring models in real time. This greatly improves the efficiency of loan approval and reduces errors and delays in manual processing.

[0062] Based on this, the present application proposes an enterprise credit assessment system, comprising a data acquisition module, a feature engineering module and an assessment module connected in sequence, wherein the data acquisition module is used to update and obtain enterprise local data and regional big data center data, and convert the regional big data center data into enterprise big data center data, the data acquisition module is respectively connected to the external data interface and the big data center to correspondingly obtain enterprise local data and regional big data center data, the data acquisition module is provided with an enterprise regional database and a conversion unit, the enterprise regional database is used to store regional related information of each enterprise, the conversion unit calls the regional related information of each enterprise from the enterprise regional database, and is used to convert the regional big data center data into enterprise big data center data;

[0063] The feature engineering module is used to mine multi-dimensional information from the enterprise's local data and the enterprise's big data center data;

[0064] The evaluation module is used to filter out important features from the multi-dimensional information input into the model, and to predict the credit score and recommended credit limit of the output enterprise.

[0065] Apply the above system to implement a corporate credit assessment method, such as Figure 1 As shown, the following steps are included:

[0066] S1. Obtain enterprise local data and regional big data center data respectively, and convert the regional big data center data into enterprise big data center data;

[0067] S2. Calculate entropy, IV and correlation coefficients for local data and big data of the enterprise to determine multi-dimensional information for modeling;

[0068] S3, perform feature screening on the multi-dimensional information input into the model, select important features from it, and construct a feature set;

[0069] S4. Use the feature set to train the XGBoost model to obtain the enterprise credit assessment model;

[0070] S5. Input the multi-dimensional information of the enterprise to be evaluated and the unified social credit code information into the enterprise credit evaluation model, and output the credit score and recommended credit limit corresponding to the enterprise to be evaluated.

[0071] In this embodiment, the construction Figure 2 The application framework shown has the following main contents:

[0072] 1. Obtaining external data such as business information, public opinion, and judicial proceedings

[0073] Collect external data such as the company's business information, public opinion data, and judicial proceedings. These data will help enrich the company's panoramic portrait and assess its credit, market reputation, and potential legal risks.

[0074] 2. Establish a building database and obtain enterprise building information

[0075] Through the SF Express address of the enterprise, we can mark it on the Amap to obtain the building information of the enterprise and establish a building database. This database will realize the correspondence between the enterprise and the building, and provide geographical support for subsequent analysis.

[0076] 3. Extract corporate tax, financial, invoice and other data from the big data center to form building-level features

[0077] The tax information, financial data and invoice records of enterprises are obtained from the big data center and summarized by building. The characteristic statistical data of enterprises in the building (such as the mean, maximum and minimum of the total tax amount) are calculated, and the relevant characteristics of the enterprises are extracted through the mapping relationship between enterprises and buildings.

[0078] 4. Clean up the enterprise data, label the character variables, and calculate the IV value to screen the features

[0079] All enterprise data are sorted and cleaned, including removing duplicates, handling missing values, etc. For character variables, label encoding is used to convert them into numerical data. Next, the IV value of each variable is calculated to screen out features with higher IV values, which have stronger distinguishing power.

[0080] 5. Use the random forest model to further screen important features

[0081] After initially screening out features with higher IV values, the random forest model is used for modeling. The random forest model can evaluate the importance of each feature to the target variable and further screen out features that contribute more to model prediction.

[0082] 6. Modeling through XGBoost model to output enterprise rating and quota

[0083] Finally, the XGBoost model is used for training. The model will predict the credit score and recommended credit limit of the enterprise based on the filtered feature set. The XGBoost model can provide more accurate results for corporate credit assessment with its efficient feature selection and powerful prediction capabilities.

[0084] Specifically, in the external data acquisition part, this embodiment uses external data acquisition channels to comprehensively obtain multi-dimensional information of the enterprise, covering industrial and commercial registration, judicial status, patent information and other aspects. First, the industrial and commercial information of the enterprise will include its basic registration information, such as the name of the enterprise, registered capital, legal representative, establishment time, business scope and shareholder structure. This information is the basic data for understanding the background, scale and business activities of the enterprise, which is helpful for evaluating the basic credit status of the enterprise.

[0085] Secondly, in terms of judicial information, we will obtain the legal records of enterprises, including litigation cases, execution information, dishonest behaviors, court judgments, etc. These judicial data can help analyze the legal risks of enterprises, especially whether the enterprises are involved in major litigation or have dishonest records, which is crucial for credit risk assessment.

[0086] Patent information demonstrates the company's innovation capabilities and technological accumulation. We will collect data on the company's patent applications, patent authorizations, and other related data to understand the company's investment and achievements in technological research and development. The number and quality of a company's patents are usually important indicators of its competitiveness and market potential.

[0087] These multi-dimensional data obtained through external procurement can provide a comprehensive portrait of the enterprise, including its operating conditions, legal compliance, risk level and technological innovation capabilities. Combined with this information, it can provide strong data support for corporate credit assessment, risk management, business cooperation decision-making, etc.

[0088] When constructing the enterprise regional database, that is, the enterprise building (park) database, first of all, through SF's address data interface, the detailed address information of each enterprise can be efficiently and accurately extracted and called. This step not only ensures the reliability and real-time nature of the address data, but also lays a solid foundation for subsequent data processing. Then, with the help of the geo-positioning function of Amap, these enterprise addresses are accurately marked on the map. Through this operation, the actual location of each enterprise can be clearly marked, and detailed information related to the building (park) can be further extracted. This includes key information such as the name of the building, the park where it is located, the building area, the floor structure, the use function, and the surrounding supporting facilities. For enterprises in the park, the database will collect an overview of the entire park, such as the area of ​​the park, the number of settled enterprises, the industrial distribution, and the operation and management of the park.

[0089] After obtaining this information, all data will be systematically organized and stored to form a comprehensive building (park) database. This database can not only display the building where a single enterprise is located and its geographical location, but also present multi-dimensional content such as the density of enterprises in the building (park), the type of building, and the industry distribution of the park. In addition, through the summary analysis of building and park information, the industrial agglomeration of enterprises in the region can be further explored, providing more in-depth insights into the business environment.

[0090] This solution constructs a corporate building database. Through the taxation, social security payment and other information in the building database, the economic contribution and development capacity of the enterprise can be further analyzed. These indicators can help the scoring model more accurately assess the financial health and social contribution of the enterprise. The building database can also contain negative information about the enterprise, such as unpaid property fees and building leasing disputes. This information helps to identify potential risk enterprises and further optimize the risk control ability of the scoring model.

[0091] When obtaining financial information of small and medium-sized enterprises, a series of key enterprise data will be obtained from the big data center, covering multiple dimensions such as tax information, invoice information, financial information, social security data, and provident fund data. These data will provide an important basis for a comprehensive understanding of the company's operating conditions, compliance level, and employee benefits. This solution obtains financial statements of small and medium-sized enterprises through the big data center channel, which can greatly improve the quality, integrity, and real-time nature of the data. This not only solves the problem of dispersion and inconsistency of multi-channel data sources in the existing technology, but also ensures that the data is updated in a timely manner and improves the prediction accuracy of the model. This method can integrate data sources such as government tax systems and online accounting software to automatically obtain financial statements of small and medium-sized enterprises, reduce manual intervention, and improve efficiency.

[0092] First of all, tax information includes the company's tax declaration, tax payment records, tax compliance, etc. By analyzing tax data, we can clearly understand the company's tax burden, tax compliance, and its performance in national and local tax policies. Tax data can also help determine the company's income, business scale and tax credit rating, which is particularly important for corporate risk assessment and credit analysis.

[0093] Invoice information is an important source of information reflecting the actual business transaction activities of an enterprise. By obtaining the invoicing information of an enterprise, including the number, amount, invoicing date and invoicing object of the invoice, it is possible to track the business transactions of the enterprise and analyze its transaction volume, customer groups and business model. Invoice information can help determine the authenticity of the enterprise's income, especially when analyzing the matching between cash flow, operating activities and tax compliance, the accuracy of invoice information is very critical.

[0094] Financial information is the core data for assessing the health of a company, covering financial statements such as the balance sheet, income statement, and cash flow statement. Through these financial statements, we can understand the profitability, debt level, cash flow, and overall financial stability of the company. Financial data can not only reflect the company's operating results, but also reveal its financial structure, capital operation, and risk tolerance, providing an important basis for credit ratings and investment decisions.

[0095] Social security data refers to the social insurance paid by enterprises for their employees, including pension, medical, unemployment, work-related injury and maternity insurance. These data can reflect the employment scale and employee welfare of enterprises. By analyzing social security payment records, it can be determined whether the enterprise provides protection for employees in accordance with national regulations, and then evaluate its compliance and employee management status.

[0096] Provident fund data shows whether the company pays housing provident fund for its employees, which is usually an important part of employee benefits. Provident fund data can not only reflect the level of employee benefits of the company, but also indirectly show the company's economic strength and the importance it attaches to its employees.

[0097] Through the integrated analysis of these multi-dimensional data, we can conduct a comprehensive assessment of the enterprise. We can see the compliance of the enterprise from the tax information, understand its business transactions and income status from the invoice information, gain insight into the economic strength and operating conditions of the enterprise through financial information, and evaluate its employee management and welfare system through social security and provident fund data. Combining these data can provide strong support for enterprise risk assessment, financial health analysis, and business decision-making.

[0098] Due to the compliance of the big data center, only building or park-level information can be obtained. According to the corresponding relationship between enterprises and buildings or parks in the enterprise regional database, the building or park data in the big data center will be converted into enterprise data. In order to more effectively utilize the massive data of the regional big data center, it will be converted through the enterprise building relationship to generate enterprise-dimensional big data center data. This conversion process is conducive to a deeper understanding of the enterprise's operating conditions and market environment, laying a solid foundation for subsequent modeling work.

[0099] Finally, when building a credit assessment model, include:

[0100] (1) Feature Engineering

[0101] By deeply mining the multi-dimensional information of regional enterprises, such as industrial and commercial information, negative records, enterprise development status, tax information, financial status, provident fund and social security information, we can fully understand the business activities and historical behaviors of enterprises. The regional scoring model uses a series of advanced feature engineering methods, including entropy method, IV value, correlation coefficient, etc., to carry out fine processing and analysis of these multi-dimensional data. Feature engineering processing is performed on the local data of enterprises and the converted big data center data, aiming to expand the data dimension and mine more valuable model input variables. These variables will cover multiple aspects such as the financial status, operational efficiency, market position, etc. of the enterprise, providing rich information support for the construction of the model.

[0102] 1. The core idea of ​​the entropy method comes from information theory, which is used to measure the information content and uncertainty of each feature in the system. The larger the entropy value, the more dispersed and uncertain the information provided by the feature; the smaller the entropy value, the higher the information concentration of the feature, and the more certain the information provided. The entropy method is usually used in multidimensional data analysis, and its importance is determined by calculating the entropy value of each feature. Features with small entropy values ​​represent that they have a higher certainty in the sample and are usually given a higher weight; features with large entropy values ​​indicate that the information they provide in the data is more widely distributed and the weight will be relatively low.

[0103] 2. When processing high-dimensional data, IV can be used to preliminarily screen many features. Generally, features with low IV values ​​(such as less than 0.02) are considered to have weak ability to distinguish the target variable and can be considered for elimination, while features with higher IV values ​​are more important. This can effectively reduce irrelevant or weakly correlated features, thereby avoiding model complexity and overfitting problems.

[0104] IV value is an important indicator to measure the contribution of each feature in distinguishing between default and non-default customers. By using IV value to analyze and select features, the model can not only effectively identify which financial indicators are most important for credit risk prediction, but also provide clear explanations on the scoring results, which enables financial institutions to adjust risk management strategies according to the contribution of each feature.

[0105] 3. Correlation coefficient analysis is used to determine the correlation between features, which can identify which variables have multicollinearity problems, avoid feature redundancy and multicollinearity, and improve the accuracy and robustness of the model.

[0106] After feature engineering is completed, a regional enterprise scoring model based on machine learning algorithms will be constructed. This model will generate a comprehensive score for each enterprise by comprehensively evaluating the multi-dimensional data of each enterprise. This score not only reflects the current operating status of the enterprise, but also predicts its future development potential and risk level.

[0107] (2) Model building

[0108] Based on multi-dimensional model input variables and with the help of advanced machine learning algorithms, regional scoring models and pre-credit line models are constructed. These two models will be able to quickly output corporate scores and pre-credit lines based on the unified social credit code as input variables, providing a basis for corporate credit assessment and financing decisions.

[0109] First, we build a random forest algorithm model. Random forest is an integrated learning method that improves the accuracy and stability of overall predictions by building multiple decision tree models. The following are its advantages in the enterprise regional scoring model: Random forest can handle a large number of features and train on different decision trees, so it is particularly suitable for the multidimensional data mentioned by users, such as business information, negative records, company development status, etc.; Random forest’s built-in feature importance calculation can help further screen out the most influential features for enterprise scoring and enhance the interpretability of the model; Since random forest uses the average results of multiple decision trees, it can well avoid the overfitting problem of a single decision tree and improve the generalization ability of the model.

[0110] After using the random forest algorithm to filter out important features, these features will be used as input to the XGBoost model. XGBoost is a decision tree algorithm based on gradient boosting, known for its efficiency and accuracy. The following is the detailed process of building an XGBoost model based on the filtered features:

[0111] Reconstruct the dataset using the filtered features: The features filtered from the random forest will form a new feature set and be used to train the XGBoost model. At this point, the dataset has been greatly streamlined, retaining only the core features that are strongly correlated with the target variable.

[0112] Training XGBoost model: Use the optimized feature set to train the XGBoost model. Unlike random forest, XGBoost uses an additive model, where each new decision tree is gradually optimized based on all previous trees. It continuously improves the accuracy of the model by minimizing the loss function, and is suitable for processing large-scale data and high-dimensional features.

[0113] The scorecard conversion formula Score=AB*ln(odds) is used to convert the probability of an enterprise becoming a high-quality enterprise predicted by the XGBoost algorithm into a score, so that different risk ratings are given to the enterprise according to the score. The enterprise pre-credit limit is predicted based on the score results and the variables processed by feature engineering.

[0114] In addition, in practical applications, a model monitoring mechanism can be established, such as tracking and analyzing model evaluation indicators every month. Once the model performance is found to be degraded or the accuracy is low, model optimization work is carried out to improve the performance, efficiency, accuracy and reliability of the machine learning model.

Claims

1. An enterprise credit evaluation system, characterized in that: It includes a data acquisition module, a feature engineering module and an evaluation module connected in sequence, wherein the data acquisition module is used to update and obtain enterprise local data and regional big data center data, and convert the regional big data center data into enterprise big data center data; The feature engineering module is used to mine input multi-dimensional information from enterprise local data and enterprise big data center data; The evaluation module is used to filter out important features from the input multi-dimensional information, and to predict and output the credit score and recommended credit limit of the enterprise.

2. The enterprise credit evaluation system according to claim 1, characterized in that: The data acquisition module is connected to the external data interface and the big data center respectively to obtain the enterprise local data and the regional big data center data accordingly.

3. The enterprise credit evaluation system according to claim 1, characterized in that: The data acquisition module is provided with an enterprise regional database and a conversion unit. The enterprise regional database is used to store the regional related information of each enterprise. The conversion unit calls the regional related information of each enterprise from the enterprise regional database to convert the regional big data center data into the enterprise big data center data.

4. A corporate credit evaluation method, applied to a corporate credit evaluation system as claimed in claim 1, characterized in that: The following steps are involved: S1. Obtain enterprise local data and regional big data center data respectively, and convert the regional big data center data into enterprise big data center data; S2. Calculate entropy, IV and correlation coefficients for local data and big data of the enterprise to determine multi-dimensional information for modeling; S3, perform feature screening on the multi-dimensional information input into the model, select important features from it, and construct a feature set; S4. Use the feature set to train the XGBoost model to obtain the enterprise credit assessment model; S5. Input the multi-dimensional information of the enterprise to be evaluated and the unified social credit code information into the enterprise credit evaluation model, and output the credit score and recommended credit limit corresponding to the enterprise to be evaluated.

5. A corporate credit assessment method according to claim 4, characterized in that: The step S1 specifically includes the following steps: S11. Obtain enterprise local data through external data collection channels; S12, constructing an enterprise regional database for storing regional related information of the enterprise; S13. Obtain the financial statement information of the enterprise from the big data center, and associate it with the regional related information of the enterprise to convert it into the enterprise big data center data accordingly.

6. A corporate credit assessment method according to claim 5, characterized in that: The enterprise local data includes the enterprise's industrial and commercial information, judicial information and patent information.

7. The enterprise credit assessment method according to claim 5, characterized in that: The financial statement information of the enterprise includes the enterprise's tax information, invoice information, financial information, social security data and provident fund data.

8. The enterprise credit assessment method according to claim 5, characterized in that: The specific process of step S12 includes: First, through the address data interface, extract and call the detailed address information of each enterprise; Then, the enterprise address information is dotted on the map, the actual location of each enterprise is marked on the map, and detailed information related to the area is extracted, including the name of the building, the park where it is located, the building area, the floor structure, the use function and the surrounding supporting facilities; For enterprises in the park, collect general information about the entire park, including the area of ​​the park, the number of settled enterprises, industrial distribution and the operation and management information of the park; Finally, all data are systematically organized and stored to construct an enterprise regional database, which is used to store the buildings where each enterprise is located and their geographical location, as well as the density of enterprises in the region, the types of buildings, and the industry distribution information of the park.

9. The enterprise credit assessment method according to claim 4, characterized in that: The specific process of step S2 includes: Entropy values ​​are calculated for enterprise local data and enterprise big data, wherein the larger the entropy value, the more dispersed and uncertain the information provided by the feature; the smaller the entropy value, the higher the information concentration of the feature, and the more certain the information provided; data information with an entropy value less than a preset first threshold is retained; Calculate the IV value for the enterprise local data and enterprise big data, and retain the data information whose IV value is higher than the preset second threshold; Conduct correlation coefficient analysis on enterprise local data and enterprise big data, and delete redundant data information with multicollinearity.

10. A method for evaluating corporate credit according to claim 4, characterized in that: The step S3 specifically uses the random forest algorithm to perform feature screening on the input multi-dimensional information to screen out the important features that have the greatest impact on the corporate credit score.

Images