Intelligent decision engine, decision system and decision method
Through the automated processing and prediction functions of the intelligent decision-making engine, the problems of low efficiency of manual audits and data silos in loan audits are solved, and more efficient and accurate loan default risk assessment and decision-making are achieved.
Patent Information
- Application Number
- CN202210301109.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-03-25
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2042-03-25
AI Technical Summary
The existing loan review process relies on manual review, which is inefficient and has serious data silos, resulting in increased credit risks and lack of Internet technology application and data sharing mechanisms, making it difficult to meet business needs.
Provides an intelligent decision-making engine that embeds the loan system through the SDK form, including initialization module, data processing module, default prediction module and model management module, and uses primary and secondary learners to build XGBoost composite tree model to automatically process and predict loan default risks. , provide decision-making opinions.
It improves the efficiency and accuracy of loan audits, reduces interference from human factors, provides a more comprehensive and objective decision-making basis, reduces credit risks, and realizes the automation of loan audits.
Smart Images

Figure CN114626938B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of computer decision systems, and in particular relates to an intelligent decision engine for loan disbursement, a decision system for loans, and a decision method for loan disbursement. Background Art
[0002] Loans are generally loans made by banks, credit unions and other institutions to organizations or individuals that use the money, and usually have specified interest rates and repayment dates. In a broad sense, loans refer to the sum of loans, discounts, overdrafts and other loan funds. Banks release the concentrated currency and monetary funds through loans, which can meet the society's demand for additional funds for expanded reproduction and promote economic development.
[0003] The general process of a loan includes: (1) the loan application stage, during which the customer needs to apply to a bank or other lending institution and fill out some application materials; (2) the loan review stage, during which the lending institution needs to review the application materials submitted by the borrower; (3) signing a loan contract. After the review is passed, the two parties will sign a loan contract, a guarantee contract, etc., depending on the specific circumstances; (4) issuing the loan.
[0004] In the entire loan process, the loan review and decision-making is the most important link. The correctness of the review decision is directly related to the credit risk issue. The existing loan process, especially the unsecured loan process, still uses manual review. The reviewer conducts a systematic inspection and investigation of the loan subject's qualifications, qualifications, credit and property status based on the credit report; in practice, for the review and decision-making of the loan process, the relevant loan reviewers often only pay attention to the identification of documents, but lack the judgment and decision-making of the overall materials and data, which can easily lead to fraud in the loan process and cause credit risks.
[0005] The existing loan approval has the following problems: (1) The penetration of Internet technology is low, the approval risk control level is uneven, and the traditional manual review is mostly adopted. There is a lack of application and assistance of Internet technology, and the review efficiency cannot meet the actual business needs; (2) The investigation data obtained by the traditional loan review cannot be connected with the approval system, the review data cannot be uniformly retained, and it is difficult to retrieve the data later; (3) Data islands are common in the loan industry, and there is a lack of sharing mechanism. It is difficult for information between financial institutions to be shared, which easily leads to omissions in the risk of loan default. Therefore, in response to the above problems, it is of great practical significance to provide an intelligent decision engine, decision system and decision method. Summary of the invention
[0006] The present invention provides an intelligent decision engine, a decision system and a decision method to solve the above problems.
[0007] To solve the above technical problems, the present invention is achieved through the following technical solutions:
[0008] An intelligent decision engine of the present invention is provided to the business end in the form of SDK and embedded in the loan system, including:
[0009] Initialization module: used to initialize the data of the entire decision engine, specifically including creating a decision engine instance and creating a model instance; the creation of the decision engine instance is to select a specific corresponding decision engine instance for subsequent decision actions of the corresponding loan personnel based on the decision engine instance; the creation of the model instance is to select a specific corresponding training model for subsequent input data training;
[0010] Data processing module: used to process the information entered by the user, including data collection and data cleaning; the data collection is used to collect the data input by the user of the loan system after interface data acquisition; the data cleaning is used to judge the correctness and standardization of the collected data, to give abnormal prompts for the data that does not meet the requirements, and to automatically modify the data into the standardized format that meets the requirements by the capture and recognition program for collection, and to automatically supplement the missing information;
[0011] Default prediction module: used to predict defaults based on new data entered by the user and output the prediction results. Before the prediction, the training model in the file is read first. The training model is trained and saved by the training module. After reading the training model, the data read by the data processing module can be passed to the default prediction module formed after training by the training model. After the prediction by the default prediction module, the prediction result is returned to the user, giving the user a judgment result on whether the loan is likely to default, and giving a decision opinion on whether to agree to lend based on the judgment result;
[0012] Model management module: used to optimize and iterate the training model in the default prediction module and manage the model version;
[0013] Decision engine database: used to provide data storage and support for the decision-making process of the decision engine. The stored data includes default prediction model information, model parameter information, model evaluation information, model prediction result information, and prediction data information.
[0014] Furthermore, the training module includes a primary learner and a secondary learner. The primary learner takes the training set and classification attributes as input of the primary learner, and then constructs an XGBoost compound tree model based on the output of the primary learner as input of the secondary learner, and finally obtains the learning model and performs prediction.
[0015] Furthermore, the primary learner includes Logistics regression, support vector machine, random forest, and neural network.
[0016] Furthermore, the workflow of the intelligent decision engine is:
[0017] Model training and corresponding evaluation result training: obtain the original data for prediction, use the data processing module to clean the original data to obtain model data that meets the requirements and standards, and then train based on the training module according to the obtained model data to obtain the corresponding training model and the model evaluation results corresponding to the training model;
[0018] Actual prediction: obtain the original user data that needs to be predicted, use the initialization module to select the corresponding training model and decision engine instance for subsequent prediction actions, use the data processing module to clean the original data to obtain model data that meets the requirements and standards, use the model data to perform actual predictions, and the default prediction module will return the prediction results to the user after prediction, giving the user a judgment result on whether the loan is likely to default, and based on the judgment result, give a decision on whether to agree to lend.
[0019] Furthermore, the original data includes basic personal information, bank credit information, unit credit information, real estate information, vehicle property information and online search information corresponding to the customer filled in by the user; the online search information is information obtained based on the customer's name, ID number, mobile phone number and work unit from the industrial and commercial information network, the court information comprehensive query network, and the criminal information query network outside the aforementioned channels.
[0020] A decision-making system, implemented by the above-mentioned intelligent decision-making engine, comprises:
[0021] Intelligent decision-making engine;
[0022] Support layer: connected to the intelligent decision-making engine, providing the intelligent decision-making engine with the data entry and access required for customer loan approval, including entering the customer's basic personal information, unit credit information, and accessing the corresponding credit system, real estate information, vehicle information, and online information platform interface, which is used for users to operate customer loan information and has a terminal with login authority verification;
[0023] Logical layer: Program code and functional structure that connects the support layer with the intelligent decision engine.
[0024] An intelligent decision-making method comprises the following steps:
[0025] S1. The user uses the terminal to log in to the decision-making system based on the authorized login authority, fills in the personal basic information and unit credit information of the customer who needs to conduct loan risk default assessment, and checks the credit system, real estate information, vehicle information and online information platform that need to be connected and captured, and automatically captures the corresponding original data information;
[0026] S2. For the original data to be predicted, the initialization module is used to select the corresponding training model and decision engine instance for subsequent prediction actions, and the data processing module is used to clean the original data to obtain model data that meets the requirements and standards, and the model data is used to perform actual prediction;
[0027] S3. After the default prediction module makes a prediction, the prediction result is returned to the user through the terminal and displayed on the terminal, giving the user a judgment result on whether the loan is likely to default, and giving a decision opinion on whether to agree to lend based on the judgment result.
[0028] Compared with the prior art, the present invention has the following beneficial effects:
[0029] The present invention predicts various types of data and information received from customers through multiple channels based on the trained model. The reference information data is wider and more comprehensive. The trained model is used to automatically give the result of the default risk based on the decision engine, rather than manually judging the default risk. The decision-making data has a wide reference range, high accuracy and efficiency, and the decision evaluation is highly objective, avoiding the interference of human factors. It is beneficial for loan review personnel to make preliminary judgments on the default risks of loan personnel accurately, efficiently, comprehensively and quickly. Moreover, it is based on computer automatic realization rather than manual evaluation, which greatly improves the efficiency of decision-making.
[0030] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS
[0031] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for describing the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.
[0032] Figure 1 A structural framework diagram of an intelligent decision-making engine of the present invention;
[0033] Figure 2 This is an architecture diagram of the decision-making system corresponding to the intelligent decision-making engine of the present invention;
[0034] Figure 3A step diagram of a decision-making method corresponding to a decision-making system according to the present invention;
[0035] Figure 4 The training steps and principle diagram corresponding to the training module of the present invention;
[0036] Figure 5 The flowchart of the model training and corresponding evaluation result training of the present invention. DETAILED DESCRIPTION
[0037] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0038] See also Figure 1 As shown, an intelligent decision engine of the present invention is provided to the business end in the form of SDK and embedded in the loan system, including:
[0039] Initialization module: used to initialize the data of the entire decision engine, including creating a decision engine instance and a model instance; creating a decision engine instance is to select a specific corresponding decision engine instance for subsequent decision actions of the corresponding loan personnel based on the decision engine instance; creating a model instance is to select a specific corresponding training model for subsequent input data training;
[0040] Data processing module: used to process the information entered by users, including data collection and data cleaning; data collection is used to collect the data input by users on the loan system end after interface data acquisition; data cleaning is used to judge the correctness and standardization of the collected data, to give abnormal prompts for data that does not meet the requirements, and to automatically modify the data into a standardized format that meets the requirements by the crawling and recognition program for collection, and to automatically supplement the missing information;
[0041] Default prediction module: used to predict defaults based on new data entered by the user and output the prediction results. Before the prediction, the training model in the file is read first. The training model is trained and saved by the training module. After reading the training model, the data read by the data processing module can be passed to the default prediction module formed after training by the training model. After the prediction by the default prediction module, the prediction result is returned to the user, giving the user a judgment result on whether the loan is likely to default, and giving a decision opinion on whether to agree to lend based on the judgment result;
[0042] Model management module: used to train new training models in the default prediction module and manage model versions;
[0043] Decision engine database: used to provide data storage and support for the decision-making process of the decision engine. The stored data includes default prediction model information, model parameter information, model evaluation information, model prediction result information, and prediction data information. The prediction model information is specifically the model file generated after model training, which is saved in the prediction model table, including the model version number, training time, parameter id, model evaluation result id, and core file. The model parameter information is specifically based on the form of a prediction model parameter table to store the model parameters required for training the model, including model parameter id, creation time, logistic regression parameter file, support vector machine parameter file, random forest parameter file, neural network parameter file, and xgboost model parameter file. The model prediction result information is specifically stored in the form of a prediction result storage table, including prediction number, data id, model version number, prediction time, and prediction result. The original data also includes basic customer information stored in the form of a customer table, including customer number, name, gender, contact number, and order number.
[0044] like Figure 4 As shown, the training module includes a primary learner and a secondary learner. The primary learner takes the training set and the classification attributes as the input of the primary learner, and then builds the XGBoost compound tree model based on the output of the primary learner as the input of the secondary learner, and finally obtains the learning model and performs prediction.
[0045] Among them, the primary learners include Logistics Regression, Support Vector Machine, Random Forest, and Neural Network.
[0046] like Figure 5 As shown, the workflow of the intelligent decision-making engine is:
[0047] Model training and corresponding evaluation result training: obtain the original data for prediction, use the data processing module to clean the original data to obtain model data that meets the requirements and standards, and then train based on the training module according to the obtained model data to obtain the corresponding training model and the model evaluation results corresponding to the training model;
[0048] Actual prediction: obtain the original user data that needs to be predicted, use the initialization module to select the corresponding training model and decision engine instance for subsequent prediction actions, use the data processing module to clean the original data to obtain model data that meets the requirements and standards, use the model data to perform actual predictions, and the default prediction module will return the prediction results to the user after prediction, giving the user a judgment result on whether the loan is likely to default, and based on the judgment result, give a decision on whether to agree to lend.
[0049] Among them, the original data includes the basic personal information, bank credit information, unit credit information, real estate information, vehicle property information and online search information corresponding to the customer filled in by the user; the online search information is information obtained based on the customer's name, ID number, mobile phone number and work unit from the Industrial and Commercial Information Network, the Court Information Comprehensive Inquiry Network, and the Criminal Information Inquiry Network outside the aforementioned channels.
[0050] like Figure 2 As shown, a decision system is implemented using the above-mentioned intelligent decision engine, including:
[0051] Intelligent decision-making engine;
[0052] Support layer: connected to the intelligent decision-making engine, providing the intelligent decision-making engine with the data entry and access required for customer loan approval, including entering the customer's basic personal information, unit credit information, and accessing the corresponding credit system, real estate information, vehicle information, and online information platform interface, which is used for users to operate customer loan information and has a terminal with login authority verification;
[0053] Logical layer: Program code and functional structure that connects the support layer with the intelligent decision engine.
[0054] like Figure 3 As shown, an intelligent decision-making method is implemented using a decision-making system as described above, comprising the following steps:
[0055] S1. The user uses the terminal to log in to the decision-making system based on the authorized login authority, fills in the personal basic information and unit credit information of the customer who needs to conduct loan risk default assessment, and checks the credit system, real estate information, vehicle information and online information platform that need to be connected and captured, and automatically captures the corresponding original data information;
[0056] S2. For the original data to be predicted, the initialization module is used to select the corresponding training model and decision engine instance for subsequent prediction actions, and the data processing module is used to clean the original data to obtain model data that meets the requirements and standards, and the model data is used to perform actual prediction;
[0057] S3. After the default prediction module makes a prediction, the prediction result is returned to the user through the terminal and displayed on the terminal, giving the user a judgment result on whether the loan is likely to default, and giving a decision opinion on whether to agree to lend based on the judgment result.
[0058] Beneficial effects:
[0059] The present invention predicts various types of data and information received from customers through multiple channels based on the trained model. The reference information data is wider and more comprehensive, and the trained model is used to automatically give the result of the default risk based on the decision engine, rather than manually judging the default risk. The decision-making data has a wide reference range, high accuracy and efficiency, and the decision evaluation is highly objective, avoiding the interference of human factors. It is conducive to loan review personnel to make preliminary judgments on the default risks of loan personnel accurately, efficiently, comprehensively and quickly, and it is based on computer automatic realization rather than manual evaluation, which greatly improves the efficiency of decision-making.
[0060] The preferred embodiments of the present invention disclosed above are only used to help illustrate the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to the specific implementation methods described. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can understand and use the present invention well. The present invention is limited only by the claims and their full scope and equivalents.
Claims
1. An intelligent system, characterized in that: include: Intelligent decision-making engine; Support layer: connected to the intelligent decision-making engine, providing the intelligent decision-making engine with the data entry and access required for customer loan approval, including entering the customer's basic personal information, unit credit information, and accessing the corresponding credit system, real estate information, vehicle information, and online information platform interface, which is used for users to operate customer loan information and has a terminal with login authority verification; Logical layer: program code and functional structure that connects the support layer with the intelligent decision engine; The intelligent decision engine is provided to the business end in the form of SDK and embedded in the loan system. The intelligent decision engine includes: Initialization module: used to initialize the data of the entire decision engine, including creating a decision engine instance and creating a model instance; the creation of the decision engine instance is to select a specific corresponding decision engine instance for subsequent decision actions of the corresponding loan personnel based on the decision engine instance; the creation of the model instance is to select a specific corresponding training model for subsequent input data training; Data processing module: used to process the information entered by the user, including data collection and data cleaning; the data collection is used to collect the data input by the user of the loan system after interface data acquisition; the data cleaning is used to judge the correctness and standardization of the collected data, to give abnormal prompts for the data that does not meet the requirements, and to automatically modify the data into the standardized format that meets the requirements by the capture and recognition program for collection, and to automatically supplement the missing information; Default prediction module: used to predict defaults based on new data entered by the user and output the prediction results. Before the prediction, the training model in the file is read first. The training model is trained and saved by the training module. After reading the training model, the data read by the data processing module can be passed to the default prediction module formed after training by the training model. After the prediction by the default prediction module, the prediction result is returned to the user, giving the user a judgment result on whether the loan is likely to default, and giving a decision opinion on whether to agree to lend based on the judgment result; Model management module: used to optimize and iterate the training model in the default prediction module and manage the model version; Decision engine database: used to provide data storage and support for the decision-making process of the decision engine. The stored data includes default prediction model information, model parameter information, model evaluation information, model prediction result information, and prediction data information.
2. An intelligent system according to claim 1, characterized in that: The training module includes a primary learner and a secondary learner. The primary learner takes the training set and classification attributes as the input of the primary learner, and then constructs an XGBoost compound tree model based on the output of the primary learner as the input of the secondary learner, and finally obtains the learning model and performs prediction.
3. An intelligent system according to claim 2, characterized in that: The primary learners include Logistics regression, support vector machine, random forest, and neural network.
4. The intelligent system according to claim 1, characterized in that: The workflow of the intelligent decision engine is: Model training and corresponding evaluation result training: obtain the original data for prediction, use the data processing module to clean the original data to obtain model data that meets the requirements and standards, and then train based on the training module according to the obtained model data to obtain the corresponding training model and the model evaluation results corresponding to the training model; Actual prediction: obtain the original user data that needs to be predicted, use the initialization module to select the corresponding training model and decision engine instance for subsequent prediction actions, use the data processing module to clean the original data to obtain model data that meets the requirements and standards, use the model data to perform actual predictions, and the default prediction module will return the prediction results to the user after prediction, giving the user a judgment result on whether the loan is likely to default, and based on the judgment result, give a decision on whether to agree to lend.
5. An intelligent system according to claim 4, characterized in that: The original data includes the basic personal information, bank credit information, unit credit information, real estate information, vehicle property information and online search information corresponding to the customer filled in by the user; the online search information is information obtained based on the customer's name, ID number, mobile phone number and work unit from the industrial and commercial information network, the court information comprehensive query network, and the criminal information query network outside the aforementioned channels.
6. An intelligent decision-making method, characterized in that: The method is implemented by an intelligent system as claimed in any one of claims 1 to 5, comprising the following steps: S1. The user uses the terminal to log in to the decision-making system based on the authorized login authority, fills in the personal basic information and unit credit information of the customer who needs to conduct loan risk default assessment, and checks the credit system, real estate information, vehicle information and online information platform that need to be connected and captured, and automatically captures the corresponding original data information; S2. For the original data to be predicted, the initialization module is used to select the corresponding training model and decision engine instance for subsequent prediction actions, and the data processing module is used to clean the original data to obtain model data that meets the requirements and standards, and the model data is used to perform actual prediction; S3. After the default prediction module makes a prediction, the prediction result is returned to the user through the terminal and displayed on the terminal, giving the user a judgment result on whether the loan is likely to default, and giving a decision opinion on whether to agree to lend based on the judgment result.
