Centralized approval method for credit extension service

Abstract

The invention relates to the technical field of credit granting data processing, and discloses a centralized approval method for credit granting services, which comprises the following steps: cleaning and classifying application data, checking whether the cleaned data is complete, judging whether the cleaned data is classified correctly, dynamically allocating approval tasks, forming an approval feedback and whole-course tracing system, optimizing an approval model, and carrying out centralized approval. According to the centralized approval method for the credit granting service, whether data omission or classification errors exist in the data preprocessing process or not is automatically judged, and under the condition that the data are inaccurate or incomplete, the data are actively recovered and adjusted; the condition that a client submits data again or an examination and approval result is inaccurate due to inaccurate or incomplete data is avoided, the examination and approval task is dynamically allocated for the application data according to the specific condition, examination and approval feedback and whole-course tracing are timely formed, the examination and approval model is timely optimized, the examination and approval process is simplified, the examination and approval time is shortened, and the examination and approval efficiency is improved. And the accuracy and objectivity of the approval result are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of credit data processing, and in particular to a centralized credit business approval method. Background Art

[0002] With the rapid development of the consumer finance industry, the traditional credit approval process has many shortcomings, mainly manifested in low approval efficiency, complex approval processes, and a lack of intelligent risk control methods. Especially when involving large amounts of customer information and complex risk assessments, the traditional decentralized approval method easily leads to inconsistent approval results and cannot effectively integrate various resources for comprehensive judgment. Therefore, a new credit business approval method is needed to achieve centralized management of the approval process, rationally allocate approval resources, and ensure scientific and consistent approval decisions.

[0003] The centralized credit approval approach aims to improve credit application processing efficiency and risk control capabilities. It defines the scope of centralized approval, establishes clear approval standards and processes, ensures transparency and operability, and builds a centralized approval information system that integrates information from various branches and departments, creating a database of customers and credit applications. The system automatically generates application numbers for credit applications submitted online. Reviewers review the information within the system and, relying on big data analysis models, conduct risk assessments on customer credit applications. A multi-level approval mechanism is established based on credit limits and risk levels. The system automatically records the approval process to ensure traceability. Approval results are fed back to applicants in real time through the system, providing an application status query function. Approved credit is dynamically monitored and potential risk changes are promptly responded to. Credit approval data is regularly analyzed to assess approval efficiency, risk control effectiveness, and customer feedback. Based on the data analysis results, approval processes and policies are continuously improved to enhance business management. Through centralized management and optimized approval processes, scientific and standardized management of credit business is achieved, enabling more efficient credit processing, effective risk control, and an improved customer experience.

[0004] The existing centralized approval method for credit business cannot determine whether there are data omissions or classification errors in the data pre-processing process that lead to inaccurate and incomplete data. It cannot restore and adjust the data when the data is inaccurate or incomplete. It cannot dynamically assign approval tasks to the data according to its specific circumstances. At the same time, it cannot form approval feedback and full traceability in a timely manner and optimize the approval model in a timely manner, resulting in complicated approval processes, long approval times, and low accuracy and objectivity of approval results. Its practicality has certain limitations. Summary of the Invention

[0005] The present invention provides a centralized credit business approval method, which has the advantages of automatically determining whether there are data omissions or classification errors during data preprocessing, and actively recovering and adjusting the data if the data is inaccurate or incomplete, thereby avoiding the situation where the customer resubmits data or the approval result is inaccurate due to inaccurate or incomplete data. The method dynamically allocates approval tasks to the application data according to its specific situation, and timely generates approval feedback and full-process traceability, and timely optimizes the approval model, thereby simplifying the approval process, shortening the approval time, and improving the accuracy and objectivity of the approval results. The method solves the problem mentioned in the above background technology that, during the credit data processing process, it is impossible to determine whether there are data omissions or classification errors during the data preprocessing, resulting in inaccurate and incomplete data, and it is impossible to recover and adjust the data if the data is inaccurate or incomplete. It is impossible to dynamically allocate approval tasks to the data according to its specific situation, and it is impossible to timely generate approval feedback and full-process traceability, and timely optimize the approval model, resulting in a complicated approval process, a long approval time, low accuracy and objectivity of the approval results, and certain limitations in its practicality.

[0006] The present invention provides the following technical solution: a centralized credit business approval method, comprising:

[0007] Obtain the target customer's application data, recorded as A;

[0008] The application data is cleaned by the cleaning function to form cleaned data, which is recorded as A':

[0009] A' = Clean(A);

[0010] The cleaned data is classified through the classification function to form classified data, which is recorded as C:

[0011] C = Classify(A');

[0012] Verify the integrity of the cleaned data and check whether there are missing values in the cleaned data A';

[0013] Get target filling data A";

[0014] Get the verification model, denoted as M;

[0015] Set up a reclassification function to reclassify the target filling data A" and generate reclassified data, recorded as C':

[0016] C'=Reclassify(A",M);

[0017] If the reclassified data C' = classified data C, and there are no missing values in the cleaned data A', then the data verification is determined to be correct, the cleaned data A' is updated, and the application data is approved;

[0018] If the reclassified data C' ≠ classified data C, or there are missing values in the cleaned data A', the data verification error is determined, the erroneous data is recorded, and the cleaned data and classified data are regenerated.

[0019] As an optional solution to the centralized credit business approval method of the present invention, the integrity verification of the cleaned data is performed to check whether there are missing values in the cleaned data A', specifically:

[0020] Get cleaned data A';

[0021] Get the format of the cleaned data A' and set it as the data format;

[0022] According to the data format, generate an analysis missing matrix, denoted as Ma:

[0023] Ma=False×shape(A');

[0024] Among them, False is the initial value of each data in the analysis missing matrix Ma, shape(●) is the data format;

[0025] By traversing the function, each element in the cleaned data A' is checked one by one to determine whether there are missing values:

[0026]

[0027] Among them, Ma i，j (·) is the ergodic function, A' i，j For each element in the cleaned data A', i and j are the row and column numbers corresponding to each element in the cleaned data A', and "" represents an empty string;

[0028] Integrate the missing values in the cleaned data A' and generate an index set, denoted as Ma indices :

[0029] Ma indices ={(i, j)|Ma i,j =True};

[0030] Integrate the analysis missing matrix Ma and index set Ma of the clean data A' indices , forming a missing data set of cleaned data A';

[0031] If there is no missing data set in the cleaned data A', it is determined that there are no missing values in the cleaned data A';

[0032] If there is a missing data set in the cleaned data A', generate filling data for the cleaned data A';

[0033] If the first filling data A" 回 ≤Second filling data A" KNN ×(1+2%), and the first filling data A" 回 ≥Second filling data A" KNN ×(1-2%), then it is determined that the missing values in the cleaned data A' can be recovered, then it is determined that there are no missing values in the cleaned data A', and the target filling data is calculated, recorded as A":

[0034]

[0035] If the first filling data A" 回 >Second filling data A" KNN ×(1+2%), or the first filling data A" 回 <Second filling data A" KNN ×(1-2%), it is determined that the missing values in the cleaned data A' cannot be recovered, and it is determined that there are missing values in the cleaned data A'.

[0036] As an optional solution of the centralized credit business approval method of the present invention, the step of generating filling data from the cleansed data A' includes generating filling data through a regression model, specifically:

[0037] Get the cleaned data A' and analyze the missing matrix Ma;

[0038] Get the regression model set model;

[0039] Through the model selection function, select the regression model of the cleaned data A', denoted as model A' :

[0040] model A' =ChooseModel(A', model);

[0041] Among them, ChooseModel(·,·) is the model selection function;

[0042] Set up a separation function to separate the columns with missing values and the columns without missing values from the data A':

[0043] features, target=PrepareData(A',Ma);

[0044] Among them, features is the feature column without missing values, target is the target column with missing values, and PrepareData(·,·) is the separation function;

[0045] Through the training function, the regression model model of the cleaned data A' A' Conduct training to form a training model, recorded as model A' _Train:

[0046] model A' _Train=Train(features, target not missing , model A' );

[0047] Among them, Train(·,·,·) is the training function, target not missing The part of non-missing values in the target column target;

[0048] Set up a prediction function to predict the missing values of the cleaned data A', denoted as predictions:

[0049] predictions=Predict(model A' _Train, missing_features);

[0050] Where Predict(·,·) is the prediction function, and missing_features is the row corresponding to the missing value in the feature column;

[0051] Set a filling function Fill(·,·,●) to fill the predicted value predictions into the missing position in the cleaned data A' to generate the first filling data, denoted as A" 回 :

[0052] A" 回 =Fill(A', Ma, predictions).

[0053] As an optional solution of the centralized credit business approval method of the present invention, the step of generating filling data from the cleansed data A' includes generating filling data through KNN filling, specifically:

[0054] Get the cleaned data A' and analyze the missing matrix Ma;

[0055] Set the number of nearest neighbors, denoted as k;

[0056] Through the feature separation function, the feature column used to calculate the similarity is separated from the cleaned data A', which is recorded as X:

[0057] X = Features(A');

[0058] Through the target separation function, the target column containing missing values is separated from the cleaned data A', denoted as Y:

[0059] Y = Target(A');

[0060] Set the similarity calculation function and calculate the similarity between each missing value in the cleaned data A' and other samples in the feature column X, recorded as distances:

[0061] distances=CalculateDistances(X i , X);

[0062] Among them, CalculateDistances(·,·) is the similarity calculation function, X i is the sample in the i-th row of the feature column X, that is, all the eigenvalues of the i-th sample;

[0063] By querying the function FindKNearest(·,·), for each missing value, we find the sample closest to it and define it as the nearest neighbor sample, denoted as k_nearest_neighbors:

[0064] k_nearest_neighbors=FindKNearest(distances, k);

[0065] Set up a prediction function to predict the missing values of the cleaned data A', recorded as:

[0066]

[0067] in,

[0068] Set a filling function Fill(·,·,·) to fill the predicted value predicted_value into the missing position in the cleaned data A' to generate the second filling data, recorded as A" KNN :

[0069] A" KNN =Fill(A',Ma,predicted_value).

[0070] As an optional solution to the centralized credit business approval method of the present invention, the recording of erroneous data and regeneration of cleaned data and classified data are specifically as follows:

[0071] If there are missing data, record the missing location and possible filling value;

[0072] If there is a classification error, record the incorrect classification and the correct classification;

[0073] Regenerate clean data and classified data based on target customers' application data;

[0074] Verify the integrity of the regenerated clean data and check whether there are missing values in the clean data A';

[0075] Generate reclassified data for the regenerated cleaned data;

[0076] If the data verification is correct, the cleansed data A' is updated and the application data is approved;

[0077] If the data verification is wrong, regenerate the cleaned data and classified data until the data verification is correct;

[0078] Get the number of times the cleaned data and classified data are regenerated, which is defined as the number of repetitions;

[0079] Set a repetition threshold;

[0080] If the number of repetitions is ≥ the repetition threshold, the application data is considered abnormal and manual review is performed.

[0081] As an optional solution to the centralized credit business approval method of the present invention, the approval of application data includes intelligent approval demand identification and automatic task allocation, specifically:

[0082] Get the risk management rules, denoted as R;

[0083] Get the cleaned data, recorded as A';

[0084] Get historical approval data, denoted as H;

[0085] Identify the application type for cleaning data by identifying the function:

[0086] T = Identify(A', R);

[0087] Through the judgment function, the approval type of the cleansing data is determined according to the application type of the cleansing data:

[0088]

[0089] If AL(T) = D, then the approval result is calculated through the primary approval function, which is recorded as D:

[0090] D = AutoDecide(A');

[0091] If AL(T) = F, the approval task is dynamically assigned by assigning the approval function:

[0092] F=Allocate(A',H).

[0093] As an optional solution to the centralized credit business approval method of the present invention, for complex applications, a comprehensive decision is generated by combining expert opinions and historical data, specifically:

[0094] Get the cleaned data, denoted as A';

[0095] Get historical approval data, denoted as H;

[0096] Obtain expert approval opinion, denoted as E;

[0097] By integrating the function, the approval result is calculated and recorded as D:

[0098] D = Integrate(A', E, H).

[0099] As an optional solution to the centralized credit business approval method of the present invention, an approval feedback and full-process traceability system is formed, specifically:

[0100] Get the cleaned data, recorded as A';

[0101] Obtain expert approval opinion, denoted as E;

[0102] Get the approval result, recorded as D;

[0103] Through the feedback function, an approval file is generated for the cleaned data to record the approval process:

[0104] F={A',E,D,Timestamp};

[0105] Among them, Timestamp is the specific date and time when the approval result D is generated.

[0106] As an optional solution to the centralized credit business approval method of the present invention, the approval model is optimized, specifically:

[0107] Get the current approval file, denoted as F;

[0108] Get the current approval result, recorded as D;

[0109] Through the evaluation function, the current approval result D is evaluated and feedback data is generated, which is recorded as F feedback :

[0110] F feedback =Evaluate(D);

[0111] Get historical approval data, denoted as H;

[0112] New historical approval data is formed, recorded as H':

[0113] H'=H∪{F};

[0114] Based on the new historical approval data H', a new approval model is formed by optimizing the function:

[0115] AM*=Optimize(H');

[0116] AM is the approval model for approving the target customer's application data, AM* is the new approval model that is optimized based on the approval model AM and the target customer's application data, and Optimize(·) is the optimization function.

[0117] The present invention has the following beneficial effects:

[0118] 1. The centralized approval method for credit business cleans the application data to form cleaned data. By checking each element in the cleaned data one by one, it is determined whether there are missing values, and thus whether there are data omissions or missing data in the application data after cleaning. If there are data omissions or missing data, the missing values are predicted by different methods to generate fill-in values. The fill-in values predicted by different methods are compared to determine whether the numerical gap is large. When the data gap is large, it means that the data prediction is not accurate enough. At this time, the application data is cleaned and checked again, and the data is actively restored and adjusted to avoid the situation where the customer resubmits the data or the approval result is inaccurate due to inaccurate or incomplete data, thereby improving the accuracy of the data and reducing the need for manual inspection of application data.

[0119] 2. The centralized approval method for credit business classifies the cleaned data before integrity verification to form classified data, and classifies the cleaned data after integrity verification to form reclassified data. If the two classified data are consistent, it means that the system classification is correct. If the two classified data are inconsistent, it means that the system classification is wrong. At this time, the application data is cleaned and checked again, and the data is actively restored and adjusted to avoid the situation where customers have to resubmit data or the approval results are inaccurate due to inaccurate or incomplete data, thereby improving data accuracy and reducing the need for manual inspection of application data.

[0120] 3. This centralized credit approval method uses data analysis results to identify the risk characteristics of applications in real time, and automatically classifies applications based on the company's risk management rules. For non-complex applications, the system automatically generates approval decisions, while for complex applications, the system uses an intelligent task allocation mechanism to assign approval tasks to corresponding decision flows. Based on machine learning models, the system automatically learns and optimizes historical approval data and decision-making effects of each strategy, dynamically assigns the decision-making process with the highest degree of matching, and generates optimized approval decisions through comprehensive analysis of multi-level and multi-dimensional data such as expert approval results, historical data, and market dynamics based on deep learning models, automatically analyzing the opinions of all experts. , risk assessment results and objective data to generate a comprehensive decision. At the same time, after the approval process is completed, the system will feedback the approval results and approval basis to the customer, and generate a complete approval file in the system in a timely manner to ensure the transparency and traceability of the approval process. After each approval, the system automatically traces back and evaluates the approval process and results, analyzes the accuracy of the approval decision and subsequent risk performance, and based on feedback data, the system continuously optimizes the approval process, decision engine and expert allocation strategy through machine learning models, optimizes the approval model in a timely manner, simplifies the approval process, shortens the approval time, improves the accuracy and objectivity of the approval results, reduces the situation of manual inspection of application data, and improves approval efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0121] Figure 1 This is a flow chart of the centralized approval method for credit business of the present invention. DETAILED DESCRIPTION

[0122] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0123] Example 1: A centralized credit approval method. Figure 1 ,include:

[0124] Obtain the target customer's application data, recorded as A;

[0125] The application data is cleaned by the cleaning function to form cleaned data, which is recorded as A':

[0126] A' = Clean(A);

[0127] The cleaned data is classified through the classification function to form classified data, which is recorded as C:

[0128] C = Classify(A');

[0129] Verify the integrity of the cleaned data and check whether there are missing values in the cleaned data A';

[0130] Get target filling data A";

[0131] Obtain a verification model, denoted as M, where the verification model M is a model used for reclassification, which can be a machine learning model (such as a decision tree, random forest, etc.) or a rule engine;

[0132] Set up a reclassification function to reclassify the target filling data A" and generate reclassified data, recorded as C':

[0133] C'=Reclassify(A",M);

[0134] If the reclassified data C' = the classified data C, and there are no missing values in the cleaned data A', then the data verification is determined to be correct, the cleaned data A' is updated, and the application data is approved. The updating of the cleaned data A' is to replace the cleaned data A' with missing values with the filled target data A", to restore the data and generate the complete cleaned data A';

[0135] If the reclassified data C' ≠ classified data C, or there are missing values in the cleaned data A', the data verification error is determined, the erroneous data is recorded, and the cleaned data and classified data are regenerated.

[0136] The integrity verification of the cleaned data is performed to check whether there are missing values in the cleaned data A', specifically:

[0137] Get cleaned data A';

[0138] Obtaining a format of the cleaned data A', which is defined as a data format, wherein the data format is a matrix shape of the cleaned data A';

[0139] According to the data format, generate an analysis missing matrix, denoted as Ma:

[0140] Ma=False×shape(A');

[0141] Among them, False is the initial value of each data in the analysis missing matrix Ma, and shape(·) is the data format;

[0142] By traversing the function, each element in the cleaned data A' is checked one by one to determine whether there are missing values:

[0143]

[0144] Among them, Mai，j (·) is the ergodic function, A' i，j For each element in the cleaned data A', i and j are the row and column numbers corresponding to each element in the cleaned data A', and "" represents an empty string;

[0145] Integrate the missing values in the cleaned data A' and generate an index set, denoted as Ma indices :

[0146] Ma indices ={(i, j)|Ma i,j =True};

[0147] Integrate the analysis missing matrix Ma and index set Ma of the clean data A' indices , forming a missing data set of cleaned data A';

[0148] If there is no missing data set in the cleaned data A', it is determined that there are no missing values in the cleaned data A';

[0149] If there is a missing data set in the cleaned data A', generate filling data for the cleaned data A';

[0150] If the first filling data A" 回 ≤Second filling data A" KNN ×(1+2%), and the first filling data A" 回 ≥Second filling data A" KNN ×(1-2%), then it is determined that the missing values in the cleaned data A' can be recovered, then it is determined that there are no missing values in the cleaned data A', and the target filling data is calculated, recorded as A":

[0151]

[0152] If the first filling data A" 回 >Second filling data A" KNN ×(1+2%), or the first filling data A" 回 <Second filling data A" KNN ×(1-2%), it is determined that the missing values in the cleaned data A' cannot be recovered, and it is determined that there are missing values in the cleaned data A'.

[0153] The generating of filling data for the cleaned data A' includes generating the filling data by filling the regression model, specifically:

[0154] Get the cleaned data A' and analyze the missing matrix Ma;

[0155] Obtain a regression model set model, wherein the regression model set model includes several regression models, such as linear regression, decision tree regression, and random forest regression;

[0156] Through the model selection function, select the regression model of the cleaned data A', denoted as model A' :

[0157] model A' =ChooseModel(A', model);

[0158] Among them, ChooseModel(·,·) is a model selection function, which can select a suitable regression model according to the characteristics of the cleaned data A' and business requirements;

[0159] Set up a separation function to separate the columns with missing values and the columns without missing values from the data A':

[0160] features, target=PrepareData(A',Ma);

[0161] Among them, features is the feature column without missing values, target is the target column with missing values, and PrepareData(·,·) is the separation function;

[0162] Through the training function, the regression model model of the cleaned data A' A' Conduct training to form a training model, recorded as model A' _Train:

[0163] model A' _Train=Train(features, target not missing , model A' );

[0164] Among them, Train(●,●,·) is the training function, target not missing The part of non-missing values in the target column target;

[0165] Set up a prediction function to predict the missing values of the cleaned data A', denoted as predictions:

[0166] predictions=Predict(model A' _Train, missing_features);

[0167] Where Predict(·,·) is the prediction function, and missing_features is the row corresponding to the missing value in the feature column;

[0168] Set a filling function Fill(·,·,·) to fill the predicted value predictions into the missing position in the cleaned data A' to generate the first filling data, denoted as A" 回 :

[0169] A" 回 =Fill(A', Ma, predictions).

[0170] This embodiment also provides that the generating of filling data for the cleaned data A' includes generating the filling data through KNN filling, specifically:

[0171] Get the cleaned data A' and analyze the missing matrix Ma;

[0172] Set a number of nearest neighbors, denoted as k, which is the number of nearest neighbor samples used to fill missing values. For example, the number of nearest neighbors is 6, where the nearest neighbor sample is the sample that is most similar to the sample where the missing value is found by the KNN filling algorithm for each missing value;

[0173] Through the feature separation function, the feature column used to calculate the similarity is separated from the cleaned data A', which is recorded as X:

[0174] X = Features(A');

[0175] Through the target separation function, the target column containing missing values is separated from the cleaned data A', denoted as Y:

[0176] Y = Target(A');

[0177] Set the similarity calculation function and calculate the similarity between each missing value in the cleaned data A' and other samples in the feature column X, recorded as distances:

[0178] distances=CalculateDistances(X i , X);

[0179] Among them, CalculateDistances(·,·) is the similarity calculation function, X i is the sample in the i-th row of the feature column X, that is, all the eigenvalues of the i-th sample;

[0180] By querying the function FindKNearest(·,·), for each missing value, we find the sample closest to it and define it as the nearest neighbor sample, denoted as k_nearest_neighbors:

[0181] k_nearest_neighbors=FindKNearest(distances, k);

[0182] Set up a prediction function to predict the missing values of the cleaned data A', recorded as:

[0183]

[0184] in,

[0185] Set a filling function Fill(·,·,·) to fill the predicted value predicted_value into the missing position in the cleaned data A' to generate the second filling data, recorded as A" KNN :

[0186] A" KNN =Fill(A',Ma,predicted_value).

[0187] Through the above method, it is automatically determined whether there are any data omissions or classification errors during the data preprocessing process. In the case of inaccurate or incomplete data, the data is actively restored and adjusted to avoid the situation where customers need to resubmit data or the approval results are inaccurate due to inaccurate or incomplete data. The approval tasks are dynamically assigned to the application data according to its specific circumstances. At the same time, approval feedback and full traceability are formed in a timely manner, and the approval model is optimized in a timely manner to simplify the approval process, shorten the approval time, and improve the accuracy and objectivity of the approval results.

[0188] Example 2: This example is an improvement made on the basis of Example 1. In the centralized credit business approval method, the recording of erroneous data and the regeneration of cleansed data and classified data are specifically as follows:

[0189] If there are missing data, record the missing location and possible filling value;

[0190] If there is a classification error, record the incorrect classification and the correct classification;

[0191] Regenerate clean data and classified data based on target customers' application data;

[0192] Verify the integrity of the regenerated clean data and check whether there are missing values in the clean data A';

[0193] Generate reclassified data for the regenerated cleaned data;

[0194] If the data verification is correct, the cleansed data A' is updated and the application data is approved;

[0195] If the data verification is wrong, regenerate the cleaned data and classified data until the data verification is correct;

[0196] Get the number of times the cleaned data and classified data are regenerated, which is defined as the number of repetitions;

[0197] Set a repetition threshold, which is 3 times;

[0198] If the number of repetitions is ≥ the repetition threshold, the application data is judged to be abnormal and manual review is performed. This means that the application data itself may have omissions, etc., and manual review is required to make a judgment.

[0199] Example 3: This example is an improvement based on Example 2. In this example, the approval of application data includes intelligent approval requirement identification and automatic task allocation, specifically:

[0200] Get the risk management rules, denoted as R;

[0201] Get the cleaned data, recorded as A';

[0202] Get historical approval data, denoted as H;

[0203] Identify the application type for cleaning data by identifying the function:

[0204] T = Identify(A', R);

[0205] Through the judgment function, the approval type of the cleansing data is determined according to the application type of the cleansing data:

[0206]

[0207] If AL(T) = D, then the approval result is calculated through the primary approval function, which is recorded as D:

[0208] D = AutoDecide(A');

[0209] If AL(T) = F, the approval task is dynamically assigned by assigning the approval function:

[0210] F=Allocate(A',H).

[0211] This embodiment also provides for generating comprehensive decisions for complex applications by combining expert opinions and historical data, specifically:

[0212] Get the cleaned data, recorded as A';

[0213] Get historical approval data, denoted as H;

[0214] Obtain expert approval opinion, denoted as E;

[0215] By integrating the function, the approval result is calculated and recorded as D:

[0216] D = Integrate(A', E, H).

[0217] This embodiment also provides for forming an approval feedback and full-process traceability system, specifically:

[0218] Get the cleaned data, recorded as A';

[0219] Obtain expert approval opinion, denoted as E;

[0220] Get the approval result, recorded as D;

[0221] Through the feedback function, an approval file is generated for the cleaned data to record the approval process:

[0222] F={A',E,D,Timestamp};

[0223] Among them, Timestamp is the specific date and time when the approval result D is generated.

[0224] This embodiment also provides an optimization of the approval model, specifically:

[0225] Get the current approval file, denoted as F;

[0226] Get the current approval result, recorded as D;

[0227] Through the evaluation function, the current approval result D is evaluated and feedback data is generated, which is recorded as F feedback :

[0228] F feedback =Evaluate(D);

[0229] Get historical approval data, denoted as H;

[0230] New historical approval data is formed, recorded as H':

[0231] H'=H∪{F};

[0232] Based on the new historical approval data H', a new approval model is formed by optimizing the function:

[0233] AM*=Optimize(H');

[0234] AM is the approval model for approving the target customer's application data, AM* is the new approval model that is optimized based on the approval model AM and the target customer's application data, and Optimize(·) is the optimization function.

[0235] This embodiment automatically determines whether there are any data omissions or classification errors during the data preprocessing process. If the data is inaccurate or incomplete, the data is actively restored and adjusted to avoid the situation where the customer resubmits data or the approval result is inaccurate due to inaccurate or incomplete data. The approval task is dynamically assigned to the application data according to its specific situation. At the same time, approval feedback and full traceability are generated in a timely manner, and the approval model is optimized in a timely manner, which simplifies the approval process, shortens the approval time, and improves the accuracy and objectivity of the approval results.

[0236] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0237] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A method for centralized credit business approval, characterized by: include: Obtain the target customer's application data, recorded as A; The application data is cleaned by the cleaning function to form cleaned data, which is recorded as A': A' = Clean(A); The cleaned data is classified through the classification function to form classified data, which is recorded as C: C = Classify(A'); Verify the integrity of the cleaned data and check whether there are missing values in the cleaned data A'; Get target filling data A"; Get the verification model, denoted as M; Set up a reclassification function to reclassify the target filling data A" and generate reclassified data, recorded as C': C'=Reclassify(A",M); If the reclassified data C' = classified data C, and there are no missing values in the cleaned data A', then the data verification is determined to be correct, the cleaned data A' is updated, and the application data is approved; If the reclassified data C' ≠ classified data C, or there are missing values in the cleaned data A', the data verification error is determined, the erroneous data is recorded, and the cleaned data and classified data are regenerated.

2. The centralized credit business approval method according to claim 1, characterized in that: The integrity verification of the cleaned data is performed to check whether there are missing values in the cleaned data A', specifically: Get cleaned data A'; Get the format of the cleaned data A' and set it as the data format; According to the data format, generate an analysis missing matrix, denoted as Ma: Ma=False×shape(A'); Among them, False is the initial value of each data in the analysis missing matrix Ma, and shape(·) is the data format; By traversing the function, each element in the cleaned data A' is checked one by one to determine whether there are missing values: Among them, Ma i，j (·) is the ergodic function, A' i，j For each element in the cleaned data A', i and j are the row and column numbers corresponding to each element in the cleaned data A', and "" represents an empty string; Integrate the missing values in the cleaned data A' and generate an index set, denoted as Ma indices : Ma indices ={(i,j)∣Ma i,j =True}; Integrate the analysis missing matrix Ma and index set Ma of the clean data A' indices , forming a missing data set of cleaned data A'; If there is no missing data set in the cleaned data A', it is determined that there are no missing values in the cleaned data A'; If there is a missing data set in the cleaned data A', generate filling data for the cleaned data A'; If the first filling data A" 回 ≤Second filling data A" KNN ×(1+2%), and the first filling data A" 回 ≥Second filling data A" KNN ×(1-2%), then it is determined that the missing values in the cleaned data A' can be recovered, then it is determined that there are no missing values in the cleaned data A', and the target filling data is calculated, recorded as A": If the first filling data A" 回 >Second filling data A" KNN ×(1+2%), or the first filling data A" 回 <Second filling data A" KNN ×(1-2%), it is determined that the missing values in the cleaned data A' cannot be recovered, and it is determined that there are missing values in the cleaned data A'.

3. The centralized credit business approval method according to claim 2, characterized in that: The generating of filling data for the cleaned data A' includes generating filling data by filling the regression model, specifically: Get the cleaned data A' and analyze the missing matrix Ma; Get the regression model set model; Through the model selection function, select the regression model of the cleaned data A', denoted as model A' : model A' =ChooseModel(A',model); Among them, ChooseModel(·,·) is the model selection function; Set up a separation function to separate the columns with missing values and the columns without missing values from the data A': features, target=PrepareData(A',Ma); Among them, features is the feature column without missing values, target is the target column with missing values, and PrepareData(·,·) is the separation function; Through the training function, the regression model model of the cleaned data A' A' Conduct training to form a training model, recorded as model A' _Train: model A' _Train=Train(features,target notmissing ,model A' ); Among them, Train(·,·,·) is the training function, target notmissing The part of non-missing values in the target column target; Set up a prediction function to predict the missing values of the cleaned data A', denoted as predictions: predictions=Predict(model A' _Train,missing_features); Where Predict(·,·) is the prediction function, and missing_features is the row corresponding to the missing value in the feature column; Set a filling function Fill(·,·,●) to fill the predicted value predictions into the missing position in the cleaned data A' to generate the first filling data, denoted as A" 回 : A" 回 =Fill(A',Ma,predictions)。 4. The centralized credit business approval method according to claim 2, characterized in that: The generating of filling data for the cleaned data A' includes generating filling data by KNN filling, specifically: Get the cleaned data A' and analyze the missing matrix Ma; Set the number of nearest neighbors, denoted as k; Through the feature separation function, the feature column used to calculate the similarity is separated from the cleaned data A', which is recorded as X: X = Features(A'); Through the target separation function, the target column containing missing values is separated from the cleaned data A', denoted as Y: Y = Target(A'); Set the similarity calculation function and calculate the similarity between each missing value in the cleaned data A' and other samples in the feature column X, recorded as distances: distances=CalculateDistances(X i ,X); Among them, CalculateDistances(●,●) is a similarity calculation function, X i is the sample in the i-th row of the feature column X, that is, all the eigenvalues of the i-th sample; By querying the function FindKNearest(●,·), for each missing value, we find the nearest sample and define it as the nearest neighbor sample, denoted as k_nearest_neighbors: k_nearest_neighbors=FindKNearest(distances, k); Set up a prediction function to predict the missing values of the cleaned data A', recorded as: in, Set a filling function Fill(●,●,●) to fill the predicted value predicted_value into the missing position in the cleaned data A' to generate the second filling data, recorded as A" KNN : A" KNN =Fill(A',Ma,predicted_value)。 5. The centralized credit business approval method according to claim 1, characterized in that: The recording of error data and regeneration of cleaned data and classified data are specifically as follows: If there are missing data, record the missing location and possible filling value; If there is a classification error, record the incorrect classification and the correct classification; Regenerate clean data and classified data based on target customers' application data; Verify the integrity of the regenerated clean data and check whether there are missing values in the clean data A'; Generate reclassified data for the regenerated cleaned data; If the data verification is correct, the cleansed data A' is updated and the application data is approved; If the data verification is wrong, regenerate the cleaned data and classified data until the data verification is correct; Get the number of times the cleaned data and classified data are regenerated, which is defined as the number of repetitions; Set a repetition threshold; If the number of repetitions is ≥ the repetition threshold, the application data is considered abnormal and manual review is performed.

6. The centralized credit business approval method according to claim 1, characterized in that: The approval of application data includes intelligent approval requirement identification and automatic task allocation, specifically: Get the risk management rules, denoted as R; Get the cleaned data, recorded as A'; Get historical approval data, denoted as H; Identify the application type for cleaning data by identifying the function: T = Identify(A', R); Through the judgment function, the approval type of the cleansing data is determined according to the application type of the cleansing data: If AL(T) = D, then the approval result is calculated through the primary approval function, which is recorded as D: D = AutoDecide(A'); If AL(T) = F, the approval task is dynamically assigned by assigning the approval function: F=Allocate(A',H).

7. The centralized credit business approval method according to claim 6, characterized in that: For complex applications, we combine expert opinions and historical data to generate comprehensive decisions, specifically: Get the cleaned data, recorded as A'; Get historical approval data, denoted as H; Obtain expert approval opinion, denoted as E; By integrating the function, the approval result is calculated and recorded as D: D = Integrate(A', E, H).

8. The centralized credit business approval method according to claims 1-7, characterized in that: Form an approval feedback and full-process traceability system, specifically: Get the cleaned data, recorded as A'; Obtain expert approval opinion, denoted as E; Get the approval result, recorded as D; Through the feedback function, an approval file is generated for the cleaned data to record the approval process: F={A',E,D,Timestamp}; Among them, Timestamp is the specific date and time when the approval result D is generated.

9. The centralized credit business approval method according to claim 8, characterized in that: Optimize the approval model, specifically: Get the current approval file, denoted as F; Get the current approval result, recorded as D; Through the evaluation function, the current approval result D is evaluated and feedback data is generated, which is recorded as F feedback : F feedback =Evaluate(D); Get historical approval data, denoted as H; New historical approval data is formed, recorded as H': H'=H∪{F}; Based on the new historical approval data H', a new approval model is formed by optimizing the function: AM*=Optimize(H'); Among them, AM is the approval model for approving the target customer's application data, AM* is the new approval model that is optimized after adding the target customer's application data to the approval model AM, and Optimize(●) is the optimization function.

Images