A bank village credit intelligent rating system and implementation method

Abstract

The application discloses a bank whole village credit granting intelligent rating system and an implementation method, belongs to the technical field of bank credit granting rating, and comprises a multi-source data acquisition module, a data preprocessing module, an intelligent rating model module, a dynamic risk control module, a rating result output module, a visual management module and a data storage module; the method is based on the system and sequentially completes system initialization, multi-source data acquisition, data preprocessing, intelligent rating model training and calculation, dynamic risk monitoring and rating adjustment, rating result output and visual management and system optimization. Through multi-dimensional data fusion, combined intelligent model and dynamic risk control mechanism, the application realizes the automation, intelligence, standardization and dynamic of whole village credit granting rating, greatly improves the rating efficiency and precision, reduces the bank credit risk, considers the credit granting fairness and inclusiveness, adapts to the rural batch credit granting scene, and helps the implementation of the rural revitalization strategy.

Description

Technical Field

[0001] This invention relates to the field of bank credit rating technology, and in particular to a smart credit rating system and implementation method for whole-village credit granting in banks. Background Technology

[0002] Whole-village credit granting is a financial service model in which banks, targeting rural areas and operating on a village-by-village basis, conduct batch credit rating and credit approval for eligible farmers and new agricultural business entities (family farms, specialized cooperatives, etc.). It is an important tool for promoting inclusive finance. Currently, the whole-village credit rating process relies heavily on manual operation, which has the following prominent technical shortcomings: 1. Inefficient rating process: Traditional whole-village credit granting requires bank staff to collect information door-to-door and fill out forms manually, followed by manual review and scoring by a dedicated person. The process is cumbersome and time-consuming. For administrative villages with a large number of farmers, it often takes several weeks or even months to complete a round of rating. This cannot meet the efficient needs of batch credit granting and is also difficult to respond quickly to farmers' financing needs.

[0003] 2. Inconsistent rating standards: Manual rating relies on the experience and judgment of the reviewers. Different reviewers have different understandings of the rating indicators and their weighting, which leads to inconsistent rating results for farmers in the same administrative village, and even situations where "the same household has different ratings". This affects the fairness and credibility of credit granting and does not meet the standardized requirements of bank risk control management.

[0004] 3. Limited data dimensions and low accuracy: Traditional rating mainly relies on a small amount of paper materials such as farmers' basic identity information and income certificates. It lacks the integration and utilization of multi-dimensional data such as farmers' credit status, production and operation, and rural governance performance. Moreover, the data collection process is prone to problems such as false information and omissions, which leads to the rating results being out of touch with farmers' actual credit level and debt repayment ability, increasing the credit risk of banks.

[0005] 4. Weak risk control capabilities: Manual rating makes it difficult to monitor and provide risk warnings in real time on massive amounts of farmer data, and it is impossible to detect changes in farmers' credit status (such as overdue payments, operating losses, etc.) in a timely manner. This leads to a lag in risk management after credit is granted, which can easily result in non-performing loans and affect the bank's asset quality.

[0006] 5. Lack of dynamic adjustment mechanism: Traditional whole-village credit rating is mostly a one-time assessment, and the rating results remain unchanged for a long time. It is impossible to dynamically optimize the rating results based on factors such as changes in farmers' production and operation, credit record updates, and adjustments in rural policies. This results in insufficient timeliness of the rating results and makes it difficult to adapt to the complex and ever-changing business environment in rural areas.

[0007] Therefore, this invention proposes an intelligent rating system and implementation method for bank village-wide credit granting that can achieve automatic data collection, intelligent analysis, standardized data, and dynamic adjustment. Summary of the Invention

[0008] To address the shortcomings of existing technologies in village-wide credit rating, such as low efficiency, inconsistent standards, limited data dimensions, weak risk control, and lack of dynamic adjustment mechanisms, this invention provides an intelligent rating system and implementation method for village-wide credit rating. By constructing a multi-source data acquisition module, an intelligent rating model, a dynamic risk control module, and a visual management module, the system achieves automation, intelligence, standardization, and dynamism in village-wide credit rating, improving rating efficiency and accuracy, reducing bank credit risk, and simultaneously ensuring inclusiveness and fairness, thus adapting to the batch scenario needs of village-wide credit rating in rural areas.

[0009] To achieve the above objectives, the present invention adopts the following technical solution: A smart rating system for whole-village credit granting by banks, S includes a multi-source data acquisition module, a data preprocessing module, a smart rating model module, a dynamic risk control module, a rating result output module, a visualization management module, and a data storage module; The multi-source data acquisition module is used to collect multi-dimensional data required for whole-village credit granting in batches through various methods such as interface calls, web crawling, manual data entry, and device linkage. This includes basic information data of farmers, credit information data of farmers, production and operation data of farmers, rural governance data, and internal bank data. The data preprocessing module is used to clean, standardize, handle missing values, and handle outliers of the collected raw data to obtain standardized data; the standardization adopts the min-max standardization method, and the formula is:

[0010] in The data is standardized, and x represents the original data. This is the minimum raw data for this indicator. The maximum original data under this indicator; the outlier handling adopts the interquartile range method, and the outlier thresholds are Q1-1.5×IQR and Q3+1.5×IQR, where IQR=Q3-Q1, Q1 is the first quartile, and Q3 is the third quartile; The intelligent rating model module includes an indicator system construction unit, a weight determination unit, a model training unit, and a rating calculation unit. The indicator system construction unit constructs a rating indicator system comprising four primary indicators—farmers' basic qualities, credit status, production and management capabilities, and rural governance performance—and several secondary indicators. The weight determination unit uses the analytic hierarchy process (AHP) to determine the weights of each indicator, obtaining weight values ​​through constructing a judgment matrix, consistency checks, and weight calculations. The model training unit uses a random forest algorithm to train the model. The rating calculation unit calculates the farmer's comprehensive credit score S based on the indicator weights and the trained model, using the following formula:

[0011] in Let i be the weight of the i-th indicator. For the standardized data of the i-th indicator, The correction coefficients for the model output are used to determine the farmer's credit rating and credit limit based on the comprehensive credit score; The dynamic risk control module is used to monitor relevant data of farmers in real time, set risk warning thresholds, trigger risk warnings, and periodically recalculate farmers' comprehensive credit scores, dynamically adjust credit ratings and credit limits, and form a risk traceability chain. The rating result output module is used to push rating results to the bank's core business system, generate rating reports, and push rating results to farmers. The visualization management module is used to display data and process status related to the credit rating of the whole village in the form of charts and graphs, and to set the operation permissions of different roles; The data storage module uses a distributed database storage system to store all data, ensuring data security, integrity, and scalability.

[0012] Furthermore, the basic information data of farmers includes farmers' names, ID numbers, family members, registered addresses, contact information, and housing conditions; the credit information data of farmers includes farmers' personal credit reports, loan delinquency records, credit card usage records, private lending information, and performance records; the production and operation data of farmers includes planting / breeding scale, agricultural product output and sales, operating years, upstream and downstream cooperative relationships, annual family income, and asset and liability status; the rural governance data includes farmers' participation in rural public welfare undertakings, performance in rural civility, village collective evaluation, and party member status and performance; the internal bank data includes farmers' historical credit records, repayment records, non-performing loan records, deposit information, and investment information.

[0013] Furthermore, the missing value handling method of the data preprocessing module is as follows: key core data is supplemented through manual entry and secondary interface calls; non-core data is filled using mean filling, median filling, or interpolation. The mean filling formula is:

[0014] in Impute missing values, where n is the number of non-missing data points for this metric. This is the i-th non-missing data point under this indicator.

[0015] Furthermore, in the weight determination unit of the intelligent rating model module, the consistency check is achieved by calculating the consistency index CI and the consistency ratio CR, where CI = ( -n) / (n-1), CR=CI / RI, where Let CR be the largest eigenvalue of the judgment matrix, n be the order of the judgment matrix, and RI be the average random consistency index. When CR < 0.1, the judgment matrix satisfies the consistency requirement.

[0016] Furthermore, the credit rating of farmers is divided into 5 levels: AAA (S≥90 points), AA (80≤S<90 points), A (70≤S<80 points), B (60≤S<70 points), and C (S<60 points), with corresponding credit limits of 300,000 yuan, 200,000 yuan, 100,000 yuan, 50,000 yuan, and 0 yuan, respectively.

[0017] A method for intelligent credit rating of entire villages by banks includes the following steps: S1: System initialization, setting relevant parameters for whole-village credit granting, connecting to external data sources, opening data collection interfaces, and completing system linkage debugging; S2: Multi-source data collection, which uses the multi-source data collection module to collect multi-dimensional data of farmers in the target administrative village in batches, and removes obviously invalid data after preliminary verification; S3: Data preprocessing involves cleaning, standardizing, handling missing values ​​and outliers in the raw data to obtain standardized data. S4: Intelligent rating model training and calculation, constructing a rating index system, determining index weights, training a random forest model, calculating farmers' comprehensive credit scores, and determining credit ratings and credit limits; S5: Dynamic risk monitoring and rating adjustment, real-time monitoring of relevant data of farmers, triggering risk warnings, periodically recalculating comprehensive credit scores, dynamically adjusting credit ratings and credit limits, forming a risk traceability chain; S6: Rating result output and visualization management, output rating results, visualize the relevant data and process status of whole village credit rating, and manage access permissions; S7: System optimization, collecting feedback, adjusting the indicator system, weights, model hyperparameters and risk warning thresholds, and optimizing system performance.

[0018] Furthermore, in step 2, for data that cannot be automatically collected through the interface, manual data entry is used to supplement the data, and the time and personnel involved in the data entry are marked.

[0019] Furthermore, in step 4, during model training, the sample data is divided into a training set and a test set in a 7:3 ratio, and accuracy, recall, and F1 score are used as model evaluation metrics until the model meets the preset accuracy requirements.

[0020] The beneficial effects of this invention are as follows: 1. Improve rating efficiency and automate batch credit granting: This invention achieves batch automatic collection of multi-dimensional data through a multi-source data acquisition module, automatic data cleaning and standardization through a data preprocessing module, and automatic calculation of credit rating through an intelligent rating model module. The entire rating process requires no manual intervention, significantly shortening the time for whole-village credit rating. For administrative villages with a large number of farmers, a round of rating can be completed within a few days, effectively solving the technical problem of low efficiency in traditional whole-village credit granting and enabling rapid response to farmers' financing needs.

[0021] 2. Unified rating standards enhance credit fairness: This invention uses the analytic hierarchy process (AHP) and random forest algorithm to construct an intelligent rating model, which clarifies the rating indicator system and weight allocation, eliminates the differences in experience-based judgments in manual rating, ensures that the rating standards for farmers within the same administrative village are unified and the results are consistent, avoids the situation of "different ratings for the same household", enhances the fairness and credibility of credit granting, and meets the standardized requirements of bank risk control management.

[0022] 3. Enriching data dimensions and improving rating accuracy: This invention integrates multi-dimensional data such as farmers' basic information, credit information, production and operation information, rural governance information, and internal bank information. Compared with traditional rating methods based on a single data dimension, it can more comprehensively and accurately reflect farmers' credit level and debt repayment ability. At the same time, through data preprocessing and combined model algorithms, it reduces the impact of false and abnormal data on the rating results.

[0023] 4. Dynamic risk control and adjustment to enhance rating timeliness: This invention features a dynamic risk control module that can monitor changes in farmers' credit status and business conditions in real time, triggering timely risk warnings. At the same time, it regularly adjusts the rating results to ensure that the rating results are consistent with the actual situation of farmers. This solves the problem of traditional rating results remaining unchanged for a long time and lacking timeliness, further reducing the risk after banks extend credit and improving the quality of bank assets.

[0024] 5. Visualized management to improve management efficiency: This invention uses a visualized management module to visually display the entire process, data distribution, and risk status of village-wide credit rating, making it easier for bank staff to quickly grasp the overall situation of village-wide credit and improve management efficiency; at the same time, it sets up access control to ensure data security and operational standards, and prevent data leakage and misoperation.

[0025] 6. Adapting to the needs of inclusive finance: This invention is designed for batch scenarios of whole-village credit granting in rural areas. It is easy to operate and has low cost, and can achieve comprehensive coverage of farmers in rural areas, especially farmers in remote areas. It effectively solves the problems of difficult credit granting and slow approval in rural areas, and promotes the extension of inclusive finance to rural areas. Attached Figure Description

[0026] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0027] Figure 1 This is a flowchart of the intelligent rating method for whole-village credit granting in this invention; Figure 2 This is a flowchart of the core data preprocessing process of the present invention. Detailed Implementation

[0028] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0029] A smart rating system for whole-village credit granting by banks includes a multi-source data acquisition module, a data preprocessing module, a smart rating model module, a dynamic risk control module, a rating result output module, a visualization management module, and a data storage module. The modules are connected in sequence to work together to achieve fully automated processing of the smart rating process for whole-village credit granting. The multi-source data acquisition module is used to collect multi-dimensional data required for whole-village credit granting, including basic information data of farmers, credit information data of farmers, production and operation data of farmers, rural governance data, and internal bank data. The multi-source data acquisition module achieves batch data collection through various methods such as interface calls, web crawling, manual data entry, and device linkage. Specifically, it includes: 1. Basic information data of farmers: including farmers' names, ID numbers, family members, registered addresses, contact information, housing conditions, etc., which are collected by calling the system interface of the public security department and village committee, and supplemented by manual entry to improve missing data; 2. Farmers' credit information data: including farmers' personal credit reports, loan delinquency records, credit card usage records, private lending information, performance records, etc., are collected through interfaces with the People's Bank of China's credit reporting system and local credit platforms, and supplemented by farmers' credit evaluation opinions provided by village committees; 3. Farmers' production and operation data: including planting / breeding scale, agricultural product output and sales, years of operation, upstream and downstream cooperation relationships, annual household income, asset and liability status, etc. This data is collected through system integration with agricultural input suppliers and agricultural product buyers, combined with IoT devices (such as temperature and humidity sensors and yield monitoring equipment), and supplemented by data reported by farmers themselves for verification. 4. Rural governance data: This includes farmers' participation in rural public welfare undertakings, rural civility (respect for the elderly, neighborly relations, straw burning, etc.), village collective evaluation, party member status and performance, etc. Data is collected through interfaces with village committees and township governments, supplemented and improved by "back-to-back" evaluation data, and referenced to relevant evaluation standards for the construction of rural credit villages. 5. Internal bank data: This includes farmers' historical credit records, repayment records, non-performing loan records, deposit information, and wealth management information, collected through interfaces of the bank's core business systems.

[0030] The data preprocessing module is used to clean, standardize, handle missing values, and handle outliers of the raw data collected by the multi-source data acquisition module, resulting in clean, standardized, and usable data to provide data support for the intelligent rating model module. The specific processing steps include: 1. Data cleaning: Remove duplicate data, spurious data (using cross-validation and logical checks), and invalid data (such as null values, garbled characters, and incorrectly formatted data) from the original data to ensure the authenticity and uniqueness of the data; 2. Data Standardization: Convert data of different formats and magnitudes to a unified standard. Use the min-max standardization method to map the data to the [0,1] interval, eliminating the influence of units. The standardization formula is as follows:

[0031] in, The data is standardized, and x represents the original data. This is the minimum raw data for this indicator. This represents the maximum raw data for this indicator; 3. Missing Value Handling: Different handling methods are used for different types of missing data. For critical core data (such as ID numbers and credit records), supplementary data is collected through manual entry and secondary API calls. For non-core data (such as some operational data), mean imputation, median imputation, or interpolation based on adjacent data are used. The imputation formula is as follows (mean imputation):

[0032] in, Impute missing values, where n is the number of non-missing data points for this metric. This is the i-th non-missing data point under this indicator; 4. Outlier Handling: Outliers are detected using the interquartile range (IQR) method. The first quartile (Q1) and third quartile (Q3) of the data are calculated, and outlier thresholds are determined as Q1 - 1.5 × IQR and Q3 + 1.5 × IQR. For outliers exceeding the thresholds, deletion, correction (replacing with the threshold), or mean correction are used to ensure the reasonableness of the data. The outlier determination formula is as follows:

[0033]

[0034] The intelligent rating model module is the core module of the system. It is used to conduct credit rating of farmers based on preprocessed standardized data. It adopts a combined model of "AHP + random forest algorithm" to ensure both scientificity and accuracy of the rating. Specifically, it includes an indicator system construction unit, a weight determination unit, a model training unit, and a rating calculation unit. 1. Indicator System Construction Unit: Construct a smart rating indicator system for whole-village credit granting, divided into primary and secondary indicators. The primary indicators include four dimensions: farmers' basic quality, credit status, production and management capabilities, and rural governance performance. Each primary indicator contains several secondary indicators, as detailed below: (1) Basic quality of farmers (first-level indicator): including family size, age structure, education level, health status, and housing conditions (second-level indicator); (2) Credit status (primary indicator): including the number of overdue payments, overdue amount, credit card usage rate, performance record, and private lending credit (secondary indicator); (3) Production and operation capacity (primary indicator): including years of operation, scale of operation, annual income, asset-liability ratio, and stability of upstream and downstream industries (secondary indicator); (4) Performance of rural governance (primary indicator): including participation in public welfare undertakings, evaluation of rural civilization, evaluation of village collective, party member status, and no record of illegal or irregular activities (secondary indicator).

[0035] 2. Weight Determination Unit: The Analytic Hierarchy Process (AHP) is used to determine the weights of each primary and secondary indicator. Through constructing a judgment matrix, performing consistency checks, and calculating weights, the weight values ​​for each indicator are obtained, ensuring the scientific and reasonable allocation of weights. The specific steps are as follows: (1) Constructing the judgment matrix: Based on the importance of each indicator, the judgment matrix A is constructed using the 1-9 scale method, where A[i][j] represents the importance of the i-th indicator relative to the j-th indicator, 1 represents equal importance, 9 represents extreme importance, and A[i][j]=1 / A[j][i]; (2) Consistency test: Calculate the largest eigenvalue of the judgment matrix. Calculate the consistency index CI and the consistency ratio CR, where CI = ( -n) / (n-1), where n is the order of the judgment matrix, CR=CI / RI (RI is the average random consistency index, which is determined according to the value of n). When CR<0.1, the judgment matrix meets the consistency requirement; otherwise, the judgment matrix is ​​readjusted. (3) Weight calculation: The eigenvalue decomposition method is used to calculate the eigenvectors of the judgment matrix, and the weight values ​​of each index are obtained after normalization. (i=1,2,...,m, where m is the total number of indicators), and satisfying .

[0036] 3. Model Training Unit: The intelligent rating model is trained using the random forest algorithm. Preprocessed standardized data is used as training samples, and human historical rating results are used as labels. By adjusting hyperparameters such as the number of decision trees, maximum depth, and node splitting threshold, the model performance is optimized and the rating accuracy is improved. During training, the sample data is divided into training and test sets in a 7:3 ratio. Accuracy, recall, and F1 score are used as model evaluation metrics to ensure the effectiveness of model training.

[0037] 4. Rating Calculation Unit: Based on the indicator weights and the trained random forest model, calculate the farmer's comprehensive credit score S. The formula for the comprehensive credit score is as follows:

[0038] in, Let i be the weight of the i-th indicator. For the standardized data of the i-th indicator, This is the correction coefficient (range [0.8, 1.2]) for the i-th index output by the random forest model, used to correct the bias in the index weights; Based on the comprehensive credit score S, farmers' credit ratings are divided into 5 levels, with the specific grading standards as follows: AAA (S≥90), AA (80≤S<90), A (70≤S<80), B (60≤S<70), and C (S<60). Among them, AAA level is granted a maximum credit line of 300,000 yuan, AA level is granted a maximum credit line of 200,000 yuan, A level is granted a maximum credit line of 100,000 yuan, B level is granted a maximum credit line of 50,000 yuan, and C level is not granted credit. This is in line with the actual needs of whole-village credit granting.

[0039] The dynamic risk control module is used to monitor the risk of farmers in real time and adjust their ratings dynamically to reduce bank credit risk. Specifically, it includes: 1. Real-time risk monitoring: Collect data on farmers' credit status, production and operation, repayment records, etc. in real time, and set risk warning thresholds (such as overdue for more than 30 days, operating income decrease of more than 50%, and illegal or irregular records). When the monitored data exceeds the warning threshold, a risk warning is automatically triggered and pushed to the bank's risk control personnel. 2. Dynamic rating adjustment: The comprehensive credit score of farmers is recalculated periodically (e.g., quarterly, semi-annually). The credit rating and credit limit of farmers are adjusted based on factors such as changes in farmers' credit status, adjustments in production and operation, and updates on rural governance performance. For farmers with overdue payments or operating losses, their credit rating and credit limit are lowered; for farmers with improved credit status and increased operating efficiency, their credit rating and credit limit are raised to ensure the timeliness and accuracy of the rating results. 3. Risk traceability: Record data, model calculation results, adjustment records, etc. throughout the farmer rating process to form a complete risk traceability chain, which facilitates verification and auditing by bank risk control personnel, and provides data support for model optimization.

[0040] The rating result output module is used to output the credit rating, comprehensive credit score, credit limit recommendation, and other results calculated by the intelligent rating model in various formats, including: 1. Push rating results to the bank's core business system for use in credit approval, loan disbursement and other business processes; 2. Generate a rating report, which includes basic information about farmers, scores for each indicator, overall score, credit rating, credit recommendations, risk warnings, etc., for bank staff and farmers to review; 3. Promote the transparency and credibility of the rating by publicizing the results through village committees, bank apps, and text messages.

[0041] The visualization management module is used to visually display and manage the entire process of village-wide credit rating, including: 1. Data visualization: Display the rating distribution of farmers in the whole village, the scores of each indicator, and the risk warning status in the form of charts (bar charts, line charts, heat maps, etc.), so that bank staff can quickly grasp the overall credit situation of the whole village; 2. Workflow visualization: Displays the progress and status of the entire process, including data collection, preprocessing, rating calculation, risk monitoring, and result output, making it easy for staff to track and manage. 3. Access Control: Set operation permissions for different roles (bank administrator, risk control personnel, village committee staff) to ensure data security and operational standards, and prevent data leakage and misoperation.

[0042] The data storage module is used to store all data, including raw data collected by the multi-source data acquisition module, standardized data processed by the data preprocessing module, training data and model parameters of the intelligent rating model, rating results, risk monitoring records, process records, etc. It uses a distributed database (such as Hadoop or MySQL cluster) for storage to ensure data security, integrity and scalability, while supporting fast data query and retrieval.

[0043] A method for intelligent rating of bank whole-village credit granting, the method being implemented based on the aforementioned intelligent rating system for bank whole-village credit granting, includes the following steps: Step 1: System initialization. Set parameters such as the scope of administrative villages for whole-village credit granting, rating indicator system, initial weight values, rating and grading standards, and risk warning thresholds. Complete the linkage debugging of each module of the system to ensure normal system operation. At the same time, connect with external data sources such as public security departments, credit reporting systems, village committees, and core business systems of banks, open data collection interfaces, and complete the configuration of data collection permissions.

[0044] Step 2: Multi-source data collection. Through the multi-source data collection module, multi-dimensional data of all farmers in the target administrative village are collected in batches, including basic information data of farmers, credit information data of farmers, production and operation data of farmers, rural governance data, and internal bank data. During the collection process, the data is initially verified and obviously invalid data is removed to ensure the initial integrity of the collected data. For data that cannot be automatically collected through the interface, it is supplemented by manual entry, and the entry time and personnel are marked for easy traceability.

[0045] Step 3: Data preprocessing. The data preprocessing module cleans, standardizes, handles missing values, and removes outliers from the collected raw data. The specific steps are as follows: 3.1 Data cleaning: Remove duplicate, false, and invalid data, and verify the authenticity of the data through cross-validation (such as ID number and name matching verification, and business data and actual production scenario verification) to ensure that the data is unique and authentic; 3.2 Data standardization: The min-max standardization method is adopted to map data of different formats and magnitudes to the [0,1] interval to eliminate the influence of units. The standardization formula is shown in formula (1). 3.3 Handling missing values: For key core data, supplementary data is collected through manual entry and secondary API calls; for non-core data, mean filling, median filling or interpolation are used to fill the missing values. The filling formula is shown in formula (2). 3.4 Outlier handling: Outliers are detected using the interquartile range (IQR) method. The outlier threshold is calculated as shown in formulas (3) and (4). Outliers exceeding the threshold are deleted, corrected, or mean-corrected to obtain standardized data.

[0046] Step 4: Intelligent rating model training and calculation. The intelligent rating model module, based on preprocessed standardized data, completes model training and farmer credit rating calculation. The specific steps are as follows: 4.1 Indicator System Construction: Through the indicator system construction unit, a whole-village credit intelligent rating indicator system is constructed, which includes 4 primary indicators and several secondary indicators; 4.2 Weight Determination: The Analytic Hierarchy Process (AHP) is used to construct a judgment matrix, perform consistency checks, and calculate the weight values ​​of each indicator. To ensure that the weight allocation is scientific and reasonable; 4.3 Model Training: The random forest algorithm is used, with standardized data as training samples and human historical rating results as labels. Hyperparameters are adjusted to train the intelligent rating model. Accuracy, recall, and F1 score are used to verify the model performance until the model meets the preset accuracy requirements. 4.4 Rating Calculation: Based on the indicator weights and the trained random forest model, the comprehensive credit score S of farmers is calculated. The calculation formula is shown in formula (5). Based on the comprehensive credit score, the credit rating and credit limit of farmers are determined.

[0047] Step 5: Dynamic Risk Monitoring and Rating Adjustment. The dynamic risk control module collects relevant data from farmers in real time, monitors risks, and periodically adjusts dynamic ratings. The specific steps are as follows: 5.1 Real-time risk monitoring: Real-time monitoring of farmers' credit status, production and operation status, repayment records, and other data. When the data exceeds the preset risk warning threshold, a risk warning is automatically triggered and pushed to the bank's risk control personnel. 5.2 Dynamic Adjustment: The comprehensive credit score of farmers is recalculated periodically (quarterly / semi-annually), and the credit rating and credit limit are adjusted in combination with changes in the farmers' circumstances, forming a dynamic adjustment record; 5.3 Risk traceability: Record all data and adjustment records throughout the rating process to form a risk traceability chain for risk control personnel to verify and audit.

[0048] Step 6: Rating Result Output and Visual Management. The rating result output module outputs the rating results in various formats, while the visual management module provides a visual display and management of the entire village credit rating process, as detailed below: 6.1 Output Results: Push the rating results to the bank's core business system, generate a rating report, and push the rating results to farmers through various channels; 6.2 Visualization: The distribution of village-wide ratings, indicator scores, risk warnings, etc., are displayed in chart form, showing the progress and status of the entire process; 6.3 Access Control: Assign operation permissions according to different roles to ensure data security and operational compliance.

[0049] Step 7: System optimization. Regularly collect feedback from bank staff and farmers. Based on the accuracy of the rating results and the effectiveness of risk warnings, adjust the rating indicator system, indicator weights, model hyperparameters, and risk warning thresholds to continuously optimize system performance and improve the accuracy and efficiency of the rating. Example

[0050] This embodiment provides a smart rating system for whole-village credit granting, applied to the whole-village credit granting business of a rural commercial bank. The specific structure is as follows: Multi-source data collection module: Connects to local public security departments, the People's Bank of China credit system, agricultural and rural affairs bureaus, village committees, and bank core business systems. It collects basic information, credit information, production and operation information, rural governance information, and internal bank information for the target administrative village (a village with 320 households) through interface calls. For some household operation data that cannot be collected through the interface (such as production data from small-scale farmers), bank staff manually supplement the data. A total of 320 households received valid data, achieving a data collection completion rate of 98%.

[0051] Data preprocessing module: The collected raw data was cleaned, removing 8 duplicate data entries and 3 false data entries (such as false income certificates); the min-max standardization method was used to map data such as farmers' annual income and business scale to the [0,1] interval; missing non-core data (such as the health status of some farmers) was supplemented by mean imputation; the interquartile range method was used to detect abnormal data, and 5 abnormal business data entries were found (such as abnormally high annual income), which were processed by threshold correction method, and finally 312 standardized data entries were obtained, with a data qualification rate of 97.5%.

[0052] The intelligent rating model module constructs a rating indicator system, with 4 primary indicators (basic farmer quality, credit status, production and management capacity, and rural governance performance) and 15 secondary indicators. The analytic hierarchy process (AHP) is used to determine the weights of each indicator, with credit status weighted at 0.35, production and management capacity at 0.3, basic farmer quality at 0.15, and rural governance performance at 0.2. A random forest algorithm is used to train the model, with 100 decision trees, a maximum depth of 8, 218 training samples (70%), and 94 test samples (30%). The model's accuracy rate is 96.8%. A comprehensive credit score is calculated for each farmer, and based on this score, 312 farmers are divided into 5 levels: AAA (28 households), AA (65 households), A (123 households), B (72 households), and C (24 households), corresponding to credit limits of 300,000 yuan, 200,000 yuan, 100,000 yuan, 50,000 yuan, and 0 yuan, respectively. The results show a 95.2% consistency with historical manual ratings.

[0053] Dynamic risk control module: Risk warning thresholds are set; warnings are triggered when overdue payments exceed 30 days or when operating income decreases by more than 50%. Real-time monitoring of farmers' repayment records and operating conditions is conducted. During this period, 3 farmers were found to be overdue (15-25 days overdue), but the warning threshold was not reached. The system automatically marked these farmers and reminded them to repay on time. Farmers' comprehensive credit scores are recalculated quarterly. 12 farmers saw their credit ratings improve due to increased operating efficiency (8 from A to AA, 4 from B to A), while 8 farmers saw their credit ratings decline due to minor overdue payments (6 from A to B, 2 from AA to A). Credit limits were adjusted promptly, effectively mitigating credit risk.

[0054] The rating result output module pushes the rating results and credit recommendations of 312 farmers to the bank's core business system, generating 312 personalized rating reports; the individual rating results are pushed to farmers through three methods: village committee bulletin board, bank APP, and SMS, with a 7-day public notice period and no objections; at the same time, the rating results are synchronized to the township government departments to provide data support for the construction of the rural credit system.

[0055] The visualization management module displays the distribution of farmers' credit ratings using bar charts, the scores of each indicator using heatmaps, and the risk warning trend using line charts. It sets up three roles: bank administrator, risk control personnel, and village committee staff, and assigns different operating permissions. Bank administrators can view the entire process data, risk control personnel can view risk warning information, and village committee staff can view the rating results of farmers in their village, ensuring standardized operation and data security.

[0056] Data storage module: All data, including raw data, standardized data, model parameters, rating results, risk records, etc., are stored using a MySQL cluster. The data storage capacity is 100GB, supporting fast data query and retrieval. The data backup cycle is 1 day to ensure data security and integrity.

[0057] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A smart rating system for whole-village credit granting by banks, characterized in that: It includes a multi-source data acquisition module, a data preprocessing module, an intelligent rating model module, a dynamic risk control module, a rating result output module, a visualization management module, and a data storage module; The multi-source data acquisition module is used to collect multi-dimensional data required for whole-village credit granting in batches through various methods such as interface calls, web crawling, manual data entry, and device linkage. This includes basic information data of farmers, credit information data of farmers, production and operation data of farmers, rural governance data, and internal bank data. The data preprocessing module is used to clean, standardize, handle missing values, and handle outliers of the collected raw data to obtain standardized data; the standardization adopts the min-max standardization method, and the formula is: in The data is standardized, and x represents the original data. This is the minimum raw data for this indicator. The maximum original data under this indicator; the outlier handling adopts the interquartile range method, and the outlier thresholds are Q1-1.5×IQR and Q3+1.5×IQR, where IQR=Q3-Q1, Q1 is the first quartile, and Q3 is the third quartile; The intelligent rating model module includes an indicator system construction unit, a weight determination unit, a model training unit, and a rating calculation unit. The indicator system construction unit constructs a rating indicator system comprising four primary indicators—farmers' basic qualities, credit status, production and management capabilities, and rural governance performance—and several secondary indicators. The weight determination unit uses the analytic hierarchy process (AHP) to determine the weights of each indicator, obtaining weight values ​​through constructing a judgment matrix, consistency checks, and weight calculations. The model training unit uses a random forest algorithm to train the model. The rating calculation unit calculates the farmer's comprehensive credit score S based on the indicator weights and the trained model, using the following formula: in Let i be the weight of the i-th indicator. For the standardized data of the i-th indicator, The correction coefficients for the model output are used to determine the farmer's credit rating and credit limit based on the comprehensive credit score; The dynamic risk control module is used to monitor relevant data of farmers in real time, set risk warning thresholds, trigger risk warnings, and periodically recalculate farmers' comprehensive credit scores, dynamically adjust credit ratings and credit limits, and form a risk traceability chain. The rating result output module is used to push rating results to the bank's core business system, generate rating reports, and push rating results to farmers. The visualization management module is used to display relevant data and process status of whole-village credit rating in chart form, and to set operation permissions for different roles; The data storage module uses a distributed database storage system to store all data, ensuring data security, integrity, and scalability.

2. The intelligent rating system for whole-village credit granting by banks according to claim 1, characterized in that, The basic information data of farmers includes farmers' names, ID numbers, family members, registered addresses, contact information, and housing conditions; the credit information data of farmers includes farmers' personal credit reports, loan delinquency records, credit card usage records, private lending information, and performance records; the production and operation data of farmers includes planting / breeding scale, agricultural product output and sales, operating years, upstream and downstream cooperative relationships, annual family income, and asset and liability status; the rural governance data includes farmers' participation in rural public welfare undertakings, performance in rural civility, village collective evaluation, and party member status and performance; the internal bank data includes farmers' historical credit records, repayment records, non-performing loan records, deposit information, and investment information.

3. The intelligent rating system for whole-village credit granting by banks according to claim 1, characterized in that, The missing value handling method of the data preprocessing module is as follows: key core data is supplemented through manual entry and secondary API calls; non-core data is filled using mean filling, median filling, or interpolation. The mean filling formula is: in Impute missing values, where n is the number of non-missing data points for this metric. This is the i-th non-missing data point under this indicator.

4. The intelligent rating system for whole-village credit granting by banks according to claim 1, characterized in that, In the weight determination unit of the intelligent rating model module, the consistency test is achieved by calculating the consistency index CI and the consistency ratio CR, where CI = ( -n) / (n-1), CR=CI / RI, where Let CR be the largest eigenvalue of the judgment matrix, n be the order of the judgment matrix, and RI be the average random consistency index. When CR < 0.1, the judgment matrix satisfies the consistency requirement.

5. The intelligent rating system for whole-village credit granting by banks according to claim 1, characterized in that, The farmers' credit rating is divided into 5 levels: AAA (S≥90 points), AA (80≤S<90 points), A (70≤S<80 points), B (60≤S<70 points), and C (S<60 points), with corresponding credit limits of 300,000 yuan, 200,000 yuan, 100,000 yuan, 50,000 yuan, and 0 yuan, respectively.

6. A method for intelligent rating of bank village-wide credit granting, characterized in that, The intelligent credit rating system for whole-village credit granting as described in any one of claims 1-5 includes the following steps: S1: System initialization, setting relevant parameters for whole-village credit granting, connecting to external data sources, opening data collection interfaces, and completing system linkage debugging; S2: Multi-source data collection, which uses the multi-source data collection module to collect multi-dimensional data of farmers in the target administrative village in batches, and removes obviously invalid data after preliminary verification; S3: Data preprocessing involves cleaning, standardizing, handling missing values ​​and outliers in the raw data to obtain standardized data. S4: Intelligent rating model training and calculation, constructing a rating index system, determining index weights, training a random forest model, calculating farmers' comprehensive credit scores, and determining credit ratings and credit limits; S5: Dynamic risk monitoring and rating adjustment, real-time monitoring of relevant data of farmers, triggering risk warnings, periodically recalculating comprehensive credit scores, dynamically adjusting credit ratings and credit limits, forming a risk traceability chain; S6: Rating result output and visualization management, output rating results, visualize the relevant data and process status of whole village credit rating, and manage access permissions; S7: System optimization, collecting feedback, adjusting the indicator system, weights, model hyperparameters and risk warning thresholds, and optimizing system performance.

7. The method for intelligent rating of whole-village credit granting by banks according to claim 6, characterized in that, In step 2, for data that cannot be automatically collected through the interface, manual data entry is used to supplement the data, and the time and personnel involved in the data entry are marked.

8. The method for intelligent rating of whole-village credit granting by banks according to claim 6, characterized in that, In step 4, during model training, the sample data is divided into training and test sets in a 7:3 ratio. Accuracy, recall, and F1 score are used as model evaluation metrics until the model meets the preset accuracy requirements.

Images