CN113919932A - Client scoring deviation detection method based on loan application scoring model - Google Patents
Client scoring deviation detection method based on loan application scoring model Download PDFInfo
- Publication number
- CN113919932A CN113919932A CN202110983469.1A CN202110983469A CN113919932A CN 113919932 A CN113919932 A CN 113919932A CN 202110983469 A CN202110983469 A CN 202110983469A CN 113919932 A CN113919932 A CN 113919932A
- Authority
- CN
- China
- Prior art keywords
- customer
- client
- group
- credit
- label
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 60
- 239000011159 matrix material Substances 0.000 claims abstract description 68
- 238000007726 management method Methods 0.000 claims description 25
- 238000012550 audit Methods 0.000 claims description 21
- 238000000034 method Methods 0.000 claims description 19
- 230000000694 effects Effects 0.000 claims description 16
- 230000008859 change Effects 0.000 claims description 12
- 238000011156 evaluation Methods 0.000 claims description 12
- 238000004458 analytical method Methods 0.000 claims description 6
- 238000010586 diagram Methods 0.000 claims description 6
- 238000012163 sequencing technique Methods 0.000 claims description 6
- 238000004364 calculation method Methods 0.000 claims description 5
- 238000012954 risk control Methods 0.000 claims description 5
- 238000007781 pre-processing Methods 0.000 claims description 4
- 238000000605 extraction Methods 0.000 claims description 3
- 230000009286 beneficial effect Effects 0.000 description 9
- 238000013461 design Methods 0.000 description 9
- 230000006870 function Effects 0.000 description 6
- 238000012986 modification Methods 0.000 description 3
- 230000004048 modification Effects 0.000 description 3
- 230000004069 differentiation Effects 0.000 description 2
- 238000013508 migration Methods 0.000 description 2
- 230000005012 migration Effects 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 230000002159 abnormal effect Effects 0.000 description 1
- 238000013475 authorization Methods 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
- 230000002194 synthesizing effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
- G06Q40/03—Credit; Loans; Processing thereof
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0631—Resource planning, allocation, distributing or scheduling for enterprises or organisations
- G06Q10/06311—Scheduling, planning or task assignment for a person or group
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0635—Risk analysis of enterprise or organisation activities
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
- G06Q10/06393—Score-carding, benchmarking or key performance indicator [KPI] analysis
Landscapes
- Business, Economics & Management (AREA)
- Human Resources & Organizations (AREA)
- Engineering & Computer Science (AREA)
- Economics (AREA)
- Strategic Management (AREA)
- Entrepreneurship & Innovation (AREA)
- Development Economics (AREA)
- Theoretical Computer Science (AREA)
- Educational Administration (AREA)
- Physics & Mathematics (AREA)
- General Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Marketing (AREA)
- Game Theory and Decision Science (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- Finance (AREA)
- Accounting & Taxation (AREA)
- Technology Law (AREA)
- Financial Or Insurance-Related Operations Such As Payment And Settlement (AREA)
Abstract
The invention provides a customer scoring deviation detection method based on a loan application scoring model, which comprises the following steps: based on the service funnel, longitudinally dividing the client group to obtain a longitudinal division result; according to the credit application month, transversely dividing the customer group to obtain a transverse division result; establishing a customer group matrix based on the longitudinal division result and the transverse division result; determining the good and bad labels of the customers in the customer group matrix based on a loan application scoring model; determining the offset degree of the customer group according to the quality discrimination of the quality labels of the customers in the customer group matrix; the invention determines the quality label and the quality discrimination of the client in the client group matrix based on the loan application scoring model, determines the offset degree of the client group, ensures the accuracy of offset detection and is convenient for risk management of the client.
Description
Technical Field
The invention relates to the technical field of data processing, in particular to a customer scoring offset detection method based on a loan application scoring model.
Background
A loan application scoring model is an important technology in the field of commercial bank loan risk control, and is used for establishing scoring for comprehensively evaluating credit of customers and giving a credit approval decision for legally collected customer authorization information by using algorithms such as statistical analysis, machine learning and the like. The credit approval is carried out by using the scores, and compared with manual approval, the credit approval method has higher efficiency, lower cost and more objective effect, and the condition that different approval personnel give different decisions can not occur.
The assessment of the loan application scoring model is mainly deduced through credit granting application, credit granting pass and withdrawal application quality labels, so the credit granting application, credit granting pass and withdrawal application division of a client are important, the client is usually divided according to the channel flow quality of the client, however, the paying client also has a large deviation with the credit granting application client after the interference of factors such as the passing of the paying client, withdrawal conversion desire, withdrawal wind control and the like, and secondly, because the credit granting wind control and the withdrawal wind control are dynamically adjusted according to bank risk preference, the paying client in each month also has a deviation, and the factors can make the deviation detection of the client inaccurate, the deviation degree of a client group cannot be well determined, and the risk management cannot be well performed on the client.
Therefore, the invention provides a customer grading deviation detection method based on a loan application grading model.
Disclosure of Invention
The invention provides a loan application scoring model-based customer scoring offset detection method, which is characterized in that based on the loan application scoring model, the quality label and the quality discrimination of a customer in a customer group matrix are determined, the offset degree of the customer group is determined, the accuracy of offset detection is ensured, and the risk management of the customer is facilitated.
The invention provides a loan application scoring model-based customer scoring offset detection method, which comprises the following steps:
step 1: based on the service funnel, longitudinally dividing the client group to obtain a longitudinal division result;
step 2: according to the credit application month, transversely dividing the customer group to obtain a transverse division result;
and step 3: establishing a customer group matrix based on the longitudinal division result and the transverse division result;
and 4, step 4: determining the good and bad labels of the customers in the customer group matrix based on a loan application scoring model;
and 5: and determining the offset degree of the customer group according to the quality discrimination of the quality labels of the customers in the customer group matrix.
In one possible way of realisation,
in step 1, based on the service funnel, longitudinally dividing the client group to obtain a longitudinally divided result, comprising:
based on the service funnel, carrying out first detection on the customer group to obtain a credit application customer;
performing secondary detection on the quality of the credit granting application client to obtain a credit granting passing client and a credit granting refusing client;
carrying out third detection on the state of the credit passing client to obtain a credit non-withdrawal client and a withdrawal application client;
and performing fourth detection on the risk control capability of the withdrawal application client to obtain withdrawal refusing clients and deposit clients and finally obtain longitudinal division results.
In one possible way of realisation,
in step 2, according to the credit application month, transversely dividing the customer group to obtain transverse division results, including:
and according to the credit granting application months of the client group, equally dividing the credit granting application months in the client group into the same group to obtain a horizontal division result.
In one possible way of realisation,
in step 3, establishing a customer group matrix based on the longitudinal division result and the transverse division result comprises:
setting a first label and a second label for the customer group by taking the longitudinal division result as a first attribute and the transverse division result as a second attribute;
establishing a matrix framework by using the longitudinal dimension of the client group matrix in the longitudinal division detection times and the transverse dimension of the client group matrix in the transverse division time period;
and determining the position of the corresponding customer in the matrix frame according to the first label and the second label of the customer group to obtain a customer group matrix.
In one possible way of realisation,
in step 4, determining the good or bad tags of the customers in the customer group matrix based on the loan application scoring model comprises:
according to the channel identification and the basic information of the customer group, evaluating the quality and the credit of the customer group, and setting a first good label and a first bad label for the customer group according to the evaluation result;
detecting the customer group based on the risk strategy of the loan application scoring model, and setting a second good label and a second bad label for the customer group according to the detection result;
and setting good labels and bad labels for the customers in the customer group based on the first good label, the first bad label and the second good label.
In one possible way of realisation,
setting a first good label and a first bad label for the customer population comprises:
acquiring channel identifications of the customer groups, and matching a plurality of target channels from a channel set based on the channel identifications;
acquiring the quality of the historical customer information of the target channel to obtain a quality analysis result;
acquiring basic information of the customer group, performing feature extraction on the basic information, and respectively acquiring first feature information, second feature information and third feature information of the basic information in three dimensions of a legal person, a company and a loan amount;
determining the credit degree and loan capacity of the customer group based on the first characteristic information and the second characteristic information;
determining a credit rating value for the customer population based on the credit rating;
comparing the loan capacity with the third characteristic information based on the loan capacity, and determining a loan capacity matching value of the customer group;
based on the credit rating value and the loan capability matching value, sequencing the customer groups in each target channel according to a preset weight calculation method to obtain a sequencing result;
setting good label passing rate for each target channel according to the quality analysis result of each target channel based on the preset good and bad label standard, selecting the customer groups in the good label passing rate from each target channel to set first good labels, and setting other labels as first bad labels.
In one possible way of realisation,
setting a second good label and a second bad label for the customer population comprises:
analyzing the risk strategy of the loan application scoring model to obtain a credit auditing standard, an asset auditing standard and a income auditing standard;
preprocessing the credit auditing standard, the asset auditing standard and the income auditing standard to obtain an auditing standard in a unified format, and establishing a credit auditing rule, an asset auditing rule and an income auditing rule by taking the auditing standard in the unified format as a reference;
dividing the credit auditing rule, the asset auditing rule and the income auditing rule respectively to obtain a plurality of credit auditing sub-rules, asset auditing sub-rules and income auditing sub-rules;
taking the credit auditing rule, the asset auditing rule and the income auditing rule as main nodes, determining an auditing node distribution diagram for sub-nodes by using the credit auditing sub-rule, the asset auditing sub-rule and the income auditing sub-rule, and establishing a label for each node and sub-node of the auditing node distribution diagram;
establishing a placement position of a rule placement model according to the checking node-label distribution map;
acquiring customer data of the customer group, and analyzing the customer data from three dimensions of credit, asset and income to obtain attributes of the customer data;
matching with the label of each node and each sub-node based on the attributes, and sequentially inputting the client data to the placement positions corresponding to the rule placement model according to the matching result for risk audit to obtain the audit value of the node or the sub-node on each placement position;
determining a weight value under each rule or sub-rule based on the risk strategy of the loan application scoring model, and generating an auditing detection model based on the weight values;
inputting the audit value of the node or the sub-node on each placement position into the audit detection model to obtain the total evaluation value of the client;
and if the total evaluation value is larger than a preset evaluation value, setting the client group as a second good label, otherwise, setting the client group as a second bad label.
In one possible way of realisation,
in step 5, determining the offset degree of the customer group according to the good-bad discrimination of the good-bad labels of the customers in the customer group matrix includes:
scoring each customer in the customer group matrix according to the loan application scoring model;
sorting each client in the client group matrix according to the scoring result, and dividing the clients in the client group matrix into n groups according to the sorting result;
calculating the KS value of the customer population matrix based on the occupation ratio of the good and bad labels of the customers in each group according to the following formula:
wherein KS represents a KS value of the customer population matrix, n represents a grouping number of the customer population matrix, FG(Scorei) Represents the ratio of the good-label clients in the ith group to all clients in the entire ith group, FB(Scorei) Representing the ratio of the bad label client in the ith group to all clients in the whole ith group;
and determining the good-bad discrimination of the good-bad labels of the customers in the customer population matrix based on the KS value, thereby determining the offset degree of the customer population.
In one possible way of realisation,
scoring each customer in the customer group matrix according to the loan application scoring model comprises:
obtaining the performance of the customer group on a transverse time axis, and obtaining the score of the customer group based on the performance;
comparing the scores of the customers with the scores of the customer groups obtained by the loan application scoring model prediction to obtain a comparison result, and judging whether the comparison result meets the preset requirement or not;
if so, taking the score of the customer group obtained by the loan application scoring model prediction as the final score of the customer group;
otherwise, determining a comparison result and a deviation value required by a preset value, resetting a trigger threshold range of the loan application scoring model according to the deviation value, recording a corresponding model effect function value in the trigger threshold range every time, and selecting a trigger threshold with the maximum model effect function value as a new trigger threshold;
obtaining loan policy adjustment information, market competition change information and operation management adjustment information in a time period from the time when the credit granting application client is shifted to the time when the credit granting application client is shifted again, and determining policy influence factors, competition influence factors and management influence factors on the basis of the loan policy adjustment information, the market competition change information and the change degree of the operation management adjustment information;
extracting loan data related to the policy influence factor, the competition influence factor and the management influence factor from a loan database respectively;
obtaining an input data set based on the loan data, inputting the input data in the input data set into the loan application scoring model, outputting a prediction result, comparing the prediction result with a scoring result corresponding to the credit application client, and determining a main influence factor;
adjusting the risk strategy of the loan application scoring model by referring to other influence factors based on the main influence factor;
and determining the final score of the customer group based on the adjusted loan application scoring model.
Additional features and advantages of the invention will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. The objectives and other advantages of the invention will be realized and attained by the structure particularly pointed out in the written description and claims hereof as well as the appended drawings.
The technical solution of the present invention is further described in detail by the accompanying drawings and embodiments.
Drawings
The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the description serve to explain the principles of the invention and not to limit the invention. In the drawings:
FIG. 1 is a flowchart of a method for detecting a deviation of a customer score based on a loan application scoring model according to an embodiment of the invention;
FIG. 2 is a flow chart of client migration in an embodiment of the present invention.
Detailed Description
The preferred embodiments of the present invention will be described in conjunction with the accompanying drawings, and it will be understood that they are described herein for the purpose of illustration and explanation and not limitation.
Example 1
The embodiment of the invention provides a loan application scoring model-based customer scoring offset detection method, which comprises the following steps of:
step 1: based on the service funnel, longitudinally dividing the client group to obtain a longitudinal division result;
step 2: according to the credit application month, transversely dividing the customer group to obtain a transverse division result;
and step 3: establishing a customer group matrix based on the longitudinal division result and the transverse division result;
and 4, step 4: determining the good and bad labels of the customers in the customer group matrix based on a loan application scoring model;
and 5: and determining the offset degree of the customer group according to the quality discrimination of the quality labels of the customers in the customer group matrix.
In this embodiment, a flow of customer migration is shown in fig. 2, and the customer group is detected for the first time based on the service funnel to obtain a trust application customer;
performing secondary detection on the quality of the credit granting application client to obtain a credit granting passing client and a credit granting refusing client;
carrying out third detection on the state of the credit passing client to obtain a credit non-withdrawal client and a withdrawal application client;
and performing fourth detection on the risk control capability of the withdrawal application client to obtain withdrawal refusing clients and deposit clients and finally obtain longitudinal division results.
In the embodiment, the customer group is longitudinally divided to obtain credit application, credit passing, credit non-withdrawal and payment customers, so that the funnel deviation condition of the customer group in a business process period is monitored, the influence of credit wind control, withdrawal willingness and withdrawal wind control on the deviation of the customer group can be detected, and guidance opinions are provided for risk management.
In this embodiment, according to the credit application month, the client groups are divided horizontally, and the obtained horizontal division result is specifically that according to a time axis (month), horizontal time axis offset detection is performed on each group of the new credit month, such as credit application, credit passing, credit non-withdrawal and loan clients, respectively, so as to find the offset degree of the same type of client groups of each month, and provide guidance opinions for risk management.
In this embodiment, determining the degree of deviation of the customer population according to the good-bad discrimination of the good-bad labels of the customers in the customer population matrix may be, for example:
drawing a longitudinal deviation detection report, firstly, comparing scoring effects on credit application and credit passing, and evaluating deviation amplitude of the passenger group influenced by the risk strategy; secondly, comparing the effects of credit passing and cash withdrawal application, the influence amplitude of the client cash withdrawal intention on the client group deviation can be evaluated; finally, the effect of the withdrawal application and the withdrawal pass is compared, and the influence amplitude of the withdrawal wind control on the deviation of the passenger groups can be evaluated.
| Type (B) | 202101 |
| Credit application | 0.30 |
| Pass through | 0.31 |
| Withdrawal application | 0.32 |
| Cash pass (put money) | 0.33 |
And drawing a transverse offset detection report. The offset amplitude of the same type of customer population on the time axis can be evaluated.
| Type (B) | 202101 | 202102 | 202103 | 202104 | 202105 | 202106 | .. |
| Pass through | 0.31 | 0.28 | 0.31 | 0.31 | 0.28 | 0.31 | 0.28 |
| Type (B) | 202101 | 202102 | 202103 | 202104 | 202105 | 202106 | .. |
| Withdrawal application | 0.32 | 0.27 | 0.32 | 0.32 | 0.27 | 0.32 | 0.27 |
And drawing a matrix type deviation detection report. And synthesizing the transverse and longitudinal deviation monitoring reports, and three-dimensionally evaluating the deviation condition of the passenger groups to provide decision reference data for risk management.
| Type (B) | 202101 | 202102 | 202103 | 202104 | 202105 | 202106 | .. |
| Credit application | 0.30 | 0.29 | 0.30 | 0.30 | 0.29 | 0.30 | 0.29 |
| Pass through | 0.31 | 0.28 | 0.31 | 0.31 | 0.28 | 0.31 | 0.28 |
| Withdrawal application | 0.32 | 0.27 | 0.32 | 0.32 | 0.27 | 0.32 | 0.27 |
| Cash pass (put money) | 0.33 | 0.25 | 0.33 | 0.33 | 0.25 | 0.33 | 0.25 |
The beneficial effect of above-mentioned design is: by horizontally and longitudinally dividing the customer group, the offset condition of the customer group in the same time point (a certain month) in the loan service flow cycle and the offset condition of the same loan service flow node (such as credit application, credit passing, credit refusal, credit non-withdrawal and payment) in the horizontal month-crossing time axis can be comprehensively and stereoscopically evaluated. The method has the advantages that the three-dimensional and objectivity of customer deviation monitoring is guaranteed, comprehensive data of customer group deviation is provided for risk management, the quality labels and the quality discrimination of customers in the customer group matrix are determined based on a loan application scoring model, the deviation degree of the customer group is determined, the accuracy of deviation detection is guaranteed, and the risk management of the customers is facilitated.
Example 2
Based on the embodiment 1, the embodiment of the invention provides a method for detecting the customer scoring deviation based on a loan application scoring model, wherein in the step 1, a customer group is longitudinally divided based on a service funnel, and the longitudinal division result is obtained by:
based on the service funnel, carrying out first detection on the customer group to obtain a credit application customer;
performing secondary detection on the quality of the credit granting application client to obtain a credit granting passing client and a credit granting refusing client;
carrying out third detection on the state of the credit passing client to obtain a credit non-withdrawal client and a withdrawal application client;
and performing fourth detection on the risk control capability of the withdrawal application client to obtain withdrawal refusing clients and deposit clients and finally obtain longitudinal division results.
The beneficial effect of above-mentioned design is: the client group is detected based on the service funnel, clients who pass the credit granting, apply for cash withdrawal and pay are obtained, accuracy of client division is guaranteed through multiple times of deviation, and a data basis is provided for detection of deviation degree of the clients.
Example 3
Based on embodiment 1, an embodiment of the present invention provides a method for detecting a customer rating deviation based on a loan application rating model, in step 2, according to a credit application month, the customer group is divided horizontally, and obtaining a horizontal division result includes:
and according to the credit granting application months of the client group, equally dividing the credit granting application months in the client group into the same group to obtain a horizontal division result.
The beneficial effect of above-mentioned design is: the client group is detected through the credit granting application month of the client group, so that the accuracy of client division is ensured, and a data basis is provided for the detection of the client deviation degree.
Example 4
Based on embodiment 1, an embodiment of the present invention provides a method for detecting a customer score offset based on a loan application score model, where in step 3, establishing a customer group matrix based on the longitudinal and transverse division results includes:
setting a first label and a second label for the customer group by taking the longitudinal division result as a first attribute and the transverse division result as a second attribute;
establishing a matrix framework by using the longitudinal dimension of the client group matrix in the longitudinal division detection times and the transverse dimension of the client group matrix in the transverse division time period;
and determining the position of the corresponding customer in the matrix frame according to the first label and the second label of the customer group to obtain a customer group matrix.
In this embodiment, the first attribute is the type of the customer group, such as credit application, credit pass, credit not withdrawal and loan customer.
In this embodiment, the second attribute is a credit application time of the customer group.
The beneficial effect of above-mentioned design is: by establishing a customer group matrix, the offset condition of the customer groups in the same time point (a certain month) in a loan service flow period and the offset condition of the same loan service flow node (such as credit application, credit passing, credit refusal, credit non-withdrawal and payment) in the same loan service flow node in a transverse cross-month time axis can be comprehensively and stereoscopically evaluated, so that the offset degree of the customer groups of the same type in each month or the customer groups of different types in the same month can be found, and guidance opinions can be provided for risk management.
Example 5
Based on embodiment 1, an embodiment of the present invention provides a method for detecting a customer score offset based on a loan application score model, and in step 4, determining a good-bad label of a customer in the customer group matrix based on the loan application score model includes:
according to the channel identification and the basic information of the customer group, evaluating the quality and the credit of the customer group, and setting a first good label and a first bad label for the customer group according to the evaluation result;
detecting the customer group based on the risk strategy of the loan application scoring model, and setting a second good label and a second bad label for the customer group according to the detection result;
and setting good labels and bad labels for the customers in the customer group based on the first good label, the first bad label and the second good label.
In this embodiment, when the first good tag and the second good tag are both of the customers, the tag of the customer is a good tag, and otherwise, the tag of the customer is a bad tag.
The beneficial effect of above-mentioned design is: by setting the quality label for each client in the client group, a basis is provided for determining the offset detection of the client group, and the offset degree of the client group is better determined.
Example 6
Based on embodiment 5, an embodiment of the present invention provides a method for detecting a customer score offset based on a loan application score model, where setting a first good tag and a first bad tag for a customer group includes:
acquiring channel identifications of the customer groups, and matching a plurality of target channels from a channel set based on the channel identifications;
acquiring the quality of the historical customer information of the target channel to obtain a quality analysis result;
acquiring basic information of the customer group, performing feature extraction on the basic information, and respectively acquiring first feature information, second feature information and third feature information of the basic information in three dimensions of a legal person, a company and a loan amount;
determining the credit degree and loan capacity of the customer group based on the first characteristic information and the second characteristic information;
determining a credit rating value for the customer population based on the credit rating;
comparing the loan capacity with the third characteristic information based on the loan capacity, and determining a loan capacity matching value of the customer group;
based on the credit rating value and the loan capability matching value, sequencing the customer groups in each target channel according to a preset weight calculation method to obtain a sequencing result;
setting good label passing rate for each target channel according to the quality analysis result of each target channel based on the preset good and bad label standard, selecting the customer groups in the good label passing rate from each target channel to set first good labels, and setting other labels as first bad labels.
In this embodiment, since the overall quality of the customer group obtained in different channels is different, the corresponding passing rate is set according to the quality of the historical customers in different target channels, and the influence of the different channels on credit granting passing can be considered, so that the obtained offset result is more in line with the actual situation.
In the embodiment, three dimensions of legal person, company and loan amount are used as indexes for whether the customer group passes, and the quality of the deviation of the customer group is ensured from multiple dimensions.
In this embodiment, the higher the credit, the greater the credit score value.
In this embodiment, the third characteristic information is within the loan capability, the higher the loan capability matching value is, and if not, the closer the third characteristic information is to the loan capability, the higher the loan capability matching value is, but the loan capability matching value is high without the third characteristic information being within the loan capability.
In this embodiment, the preset weight calculation method may assign different weights to the credit rating value and the loan capability matching value according to the actual credit granting requirement, so as to obtain the final result, for example, the weight set for the credit rating value is 0.8, and the weight set for the loan capability matching value is 0.6.
In this embodiment, the credit granting application clients in each target channel are ranked in descending order according to the final result obtained by the preset weight calculation method.
The beneficial effect of above-mentioned design is: by setting different good label passing rates under different target channels, the accuracy of setting the first good label is guaranteed, and the deviation degree of a customer group is better determined.
Example 7
Based on embodiment 5, an embodiment of the present invention provides a method for detecting a customer score deviation based on a loan application score model, where setting a second good tag and a second bad tag for the customer group includes:
analyzing the risk strategy of the loan application scoring model to obtain a credit auditing standard, an asset auditing standard and a income auditing standard;
preprocessing the credit auditing standard, the asset auditing standard and the income auditing standard to obtain an auditing standard in a unified format, and establishing a credit auditing rule, an asset auditing rule and an income auditing rule by taking the auditing standard in the unified format as a reference;
dividing the credit auditing rule, the asset auditing rule and the income auditing rule respectively to obtain a plurality of credit auditing sub-rules, asset auditing sub-rules and income auditing sub-rules;
taking the credit auditing rule, the asset auditing rule and the income auditing rule as main nodes, determining an auditing node distribution diagram for sub-nodes by using the credit auditing sub-rule, the asset auditing sub-rule and the income auditing sub-rule, and establishing a label for each node and sub-node of the auditing node distribution diagram;
establishing a placement position of a rule placement model according to the checking node-label distribution map;
acquiring customer data of the customer group, and analyzing the customer data from three dimensions of credit, asset and income to obtain attributes of the customer data;
matching with the label of each node and each sub-node based on the attributes, and sequentially inputting the client data to the placement positions corresponding to the rule placement model according to the matching result for risk audit to obtain the audit value of the node or the sub-node on each placement position;
determining a weight value under each rule or sub-rule based on the risk strategy of the loan application scoring model, and generating an auditing detection model based on the weight values;
inputting the audit value of the node or the sub-node on each placement position into the audit detection model to obtain the total evaluation value of the client;
and if the total evaluation value is larger than a preset evaluation value, setting the client group as a second good label, otherwise, setting the client group as a second bad label.
In this embodiment, the preprocessing is standardized for credit audit standards, asset audit standards, revenue audit standards.
In this embodiment, the unified format of the audit criteria facilitates the establishment of the rules.
In this embodiment, the credit auditing sub-rule includes a personal credit detection sub-rule, a family member credit detection sub-rule, and a company credit detection sub-rule.
In this embodiment, the asset audit sub-rule includes a house property detection sub-rule, an automobile asset detection sub-rule, and a company asset detection sub-rule.
In this embodiment, the income checking sub-rule includes a work stability detection sub-rule and other income detection sub-rules.
In this embodiment, the rule placement model establishes a plurality of data receiving and placing positions from different dimensions, each position is provided with a corresponding auditing method, and when client data is received, client data is audited to obtain a corresponding auditing value.
In this embodiment, the weight value under each rule or sub-rule is used to indicate the importance degree of the corresponding rule or sub-rule, and the more important the importance is, the larger the value is.
In this embodiment, generating the audit detection model based on the weight value includes performing weighting processing on each rule position under the audit detection model by using the audit value to obtain a new audit rule.
The beneficial effect of above-mentioned design is: and setting a second good-bad label of the customer group through a risk strategy of the loan application scoring model, and setting the second good-bad label for the customer group from the aspect of the risk strategy to ensure that the deviation degree of the customer group is better determined.
Example 8
Based on embodiment 9, an embodiment of the present invention provides a method for detecting a deviation of a customer score based on a loan application scoring model, where, in step 5, determining a deviation degree of a customer group according to a good-bad discrimination degree of a good-bad label of the customer in a customer group matrix includes:
scoring each customer in the customer group matrix according to the loan application scoring model;
sorting each client in the client group matrix according to the scoring result, and dividing the clients in the client group matrix into n groups according to the sorting result;
calculating the KS value of the customer population matrix based on the occupation ratio of the good and bad labels of the customers in each group according to the following formula:
wherein KS represents a KS value of the customer population matrix, n represents a grouping number of the customer population matrix, FG(Scorei) Represents the ratio of the good-label clients in the ith group to all clients in the entire ith group, FB(Scorei) Representing the ratio of the bad label client in the ith group to all clients in the whole ith group;
and determining the good-bad discrimination of the good-bad labels of the customers in the customer population matrix based on the KS value, thereby determining the offset degree of the customer population.
In this embodiment, the KS value of the customer population matrix is used to evaluate the degree of discrimination of the application score on good-label customers and bad-label customers, and the greater the degree of discrimination, the stronger the risk ranking capability of the score, and the better the effect.
The beneficial effect of above-mentioned design is: and determining the good-bad discrimination of the client group matrix through the KS value index so as to determine the offset degree of the client group, ensure the accuracy of offset detection and facilitate the risk management of the client.
Example 9
Based on embodiment 8, an embodiment of the present invention provides a method for detecting a customer score offset based on a loan application scoring model, where scoring each customer in the customer group matrix according to the loan application scoring model includes:
obtaining the performance of the customer group on a transverse time axis, and obtaining the score of the customer group based on the performance;
comparing the scores of the customers with the scores of the customer groups obtained by the loan application scoring model prediction to obtain a comparison result, and judging whether the comparison result meets the preset requirement or not;
if so, taking the score of the customer group obtained by the loan application scoring model prediction as the final score of the customer group;
otherwise, determining a comparison result and a deviation value required by a preset value, resetting a trigger threshold range of the loan application scoring model according to the deviation value, recording a corresponding model effect function value in the trigger threshold range every time, and selecting a trigger threshold with the maximum model effect function value as a new trigger threshold;
obtaining loan policy adjustment information, market competition change information and operation management adjustment information in a time period from the time when the credit granting application client is shifted to the time when the credit granting application client is shifted again, and determining policy influence factors, competition influence factors and management influence factors on the basis of the loan policy adjustment information, the market competition change information and the change degree of the operation management adjustment information;
extracting loan data related to the policy influence factor, the competition influence factor and the management influence factor from a loan database respectively;
obtaining an input data set based on the loan data, inputting the input data in the input data set into the loan application scoring model, outputting a prediction result, comparing the prediction result with a scoring result corresponding to the credit application client, and determining a main influence factor;
adjusting the risk strategy of the loan application scoring model by referring to other influence factors based on the main influence factor;
and determining the final score of the customer group based on the adjusted loan application scoring model.
In this embodiment, the trigger threshold represents the differentiation capability of the loan application scoring model, for example, if the trigger value is smaller than the trigger threshold range, it indicates that the loan application scoring model has no discrimination capability, if the trigger value is within the trigger threshold range, it indicates that the loan application scoring model has the differentiation capability, and if the trigger value is larger than the trigger threshold range, it indicates that the loan application scoring model is abnormal.
In this embodiment, the model effect function value is used to represent the scoring effect of the loan application scoring model, and the value of the model effect function value is larger as the effect is better.
In this embodiment, the greater the change degree of the loan policy adjustment information, the market competition change information, and the operation management adjustment information is, the greater the corresponding policy influence factor, competition influence factor, and management influence factor are.
In this embodiment, the loan data is divided into three groups, which correspond to the policy impact factor, the competition impact factor, and the management impact factor.
In this embodiment, the main influence factor is the largest influence factor on the loan application scoring model, and it is shown that the larger the difference between the prediction result and the scoring result corresponding to the credit application client is, the corresponding influence factor is the main influence factor.
The beneficial effect of above-mentioned design is: whether the strategy of the loan application scoring model needs to be adjusted is determined according to the deviation between the performance of a customer group on a time axis and the prediction result of the loan application scoring model, and the risk strategy of the loan application scoring model is adjusted according to loan policy adjustment information, market competition change information and operation management adjustment information in the time period from the time when a credit granting application customer is deviated to the time when the credit granting application customer is deviated again, so that the loan application scoring model can be adjusted timely according to the actual situation, the scoring effect accuracy is improved, and the scoring accuracy of the customer group is ensured.
It will be apparent to those skilled in the art that various changes and modifications may be made in the present invention without departing from the spirit and scope of the invention. Thus, if such modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include such modifications and variations.
Claims (9)
1. A customer scoring deviation detection method based on a loan application scoring model is characterized by comprising the following steps:
step 1: based on the service funnel, longitudinally dividing the client group to obtain a longitudinal division result;
step 2: according to the credit application month, transversely dividing the customer group to obtain a transverse division result;
and step 3: establishing a customer group matrix based on the longitudinal division result and the transverse division result;
and 4, step 4: determining the good and bad labels of the customers in the customer group matrix based on a loan application scoring model;
and 5: and determining the offset degree of the customer group according to the quality discrimination of the quality labels of the customers in the customer group matrix.
2. The method as claimed in claim 1, wherein in step 1, the customer group is divided longitudinally based on the service funnel, and the obtaining of the longitudinal division result includes:
based on the service funnel, carrying out first detection on the customer group to obtain a credit application customer;
performing secondary detection on the quality of the credit granting application client to obtain a credit granting passing client and a credit granting refusing client;
carrying out third detection on the state of the credit passing client to obtain a credit non-withdrawal client and a withdrawal application client;
and performing fourth detection on the risk control capability of the withdrawal application client to obtain withdrawal refusing clients and deposit clients and finally obtain longitudinal division results.
3. The method as claimed in claim 1, wherein in step 2, the customer group is divided horizontally according to the trust application month, and obtaining the horizontal division result comprises:
and according to the credit granting application months of the client group, equally dividing the credit granting application months in the client group into the same group to obtain a horizontal division result.
4. The method as claimed in claim 1, wherein the step 3 of establishing the customer group matrix based on the longitudinal and transverse division results comprises:
setting a first label and a second label for the customer group by taking the longitudinal division result as a first attribute and the transverse division result as a second attribute;
establishing a matrix framework by using the longitudinal dimension of the client group matrix in the longitudinal division detection times and the transverse dimension of the client group matrix in the transverse division time period;
and determining the position of the corresponding customer in the matrix frame according to the first label and the second label of the customer group to obtain a customer group matrix.
5. The method as claimed in claim 1, wherein the step 4 of determining the good or bad tags of the clients in the client group matrix based on the loan application scoring model comprises:
according to the channel identification and the basic information of the customer group, evaluating the quality and the credit of the customer group, and setting a first good label and a first bad label for the customer group according to the evaluation result;
detecting the customer group based on the risk strategy of the loan application scoring model, and setting a second good label and a second bad label for the customer group according to the detection result;
and setting good labels and bad labels for the customers in the customer group based on the first good label, the first bad label and the second good label.
6. The loan application scoring model-based customer scoring offset detection method according to claim 5, wherein setting a first good label and a first bad label for the customer group comprises:
acquiring channel identifications of the customer groups, and matching a plurality of target channels from a channel set based on the channel identifications;
acquiring the quality of the historical customer information of the target channel to obtain a quality analysis result;
acquiring basic information of the customer group, performing feature extraction on the basic information, and respectively acquiring first feature information, second feature information and third feature information of the basic information in three dimensions of a legal person, a company and a loan amount;
determining the credit degree and loan capacity of the customer group based on the first characteristic information and the second characteristic information;
determining a credit rating value for the customer population based on the credit rating;
comparing the loan capacity with the third characteristic information based on the loan capacity, and determining a loan capacity matching value of the customer group;
based on the credit rating value and the loan capability matching value, sequencing the customer groups in each target channel according to a preset weight calculation method to obtain a sequencing result;
setting good label passing rate for each target channel according to the quality analysis result of each target channel based on the preset good and bad label standard, selecting the customer groups in the good label passing rate from each target channel to set first good labels, and setting other labels as first bad labels.
7. The loan application scoring model-based customer scoring offset detection method according to claim 5, wherein setting a second good label and a second bad label for the customer group comprises:
analyzing the risk strategy of the loan application scoring model to obtain a credit auditing standard, an asset auditing standard and a income auditing standard;
preprocessing the credit auditing standard, the asset auditing standard and the income auditing standard to obtain an auditing standard in a unified format, and establishing a credit auditing rule, an asset auditing rule and an income auditing rule by taking the auditing standard in the unified format as a reference;
dividing the credit auditing rule, the asset auditing rule and the income auditing rule respectively to obtain a plurality of credit auditing sub-rules, asset auditing sub-rules and income auditing sub-rules;
taking the credit auditing rule, the asset auditing rule and the income auditing rule as main nodes, determining an auditing node distribution diagram for sub-nodes by using the credit auditing sub-rule, the asset auditing sub-rule and the income auditing sub-rule, and establishing a label for each node and sub-node of the auditing node distribution diagram;
establishing a placement position of a rule placement model according to the checking node-label distribution map;
acquiring customer data of the customer group, and analyzing the customer data from three dimensions of credit, asset and income to obtain attributes of the customer data;
matching with the label of each node and each sub-node based on the attributes, and sequentially inputting the client data to the placement positions corresponding to the rule placement model according to the matching result for risk audit to obtain the audit value of the node or the sub-node on each placement position;
determining a weight value under each rule or sub-rule based on the risk strategy of the loan application scoring model, and generating an auditing detection model based on the weight values;
inputting the audit value of the node or the sub-node on each placement position into the audit detection model to obtain the total evaluation value of the client;
and if the total evaluation value is larger than a preset evaluation value, setting the client group as a second good label, otherwise, setting the client group as a second bad label.
8. The method as claimed in claim 1, wherein the step 5 of determining the deviation degree of the customer group according to the quality degree of the quality label of the customer in the customer group matrix comprises:
scoring each customer in the customer group matrix according to the loan application scoring model;
sorting each client in the client group matrix according to the scoring result, and dividing the clients in the client group matrix into n groups according to the sorting result;
calculating the KS value of the customer population matrix based on the occupation ratio of the good and bad labels of the customers in each group according to the following formula:
wherein KS represents a KS value of the customer population matrix, n represents a grouping number of the customer population matrix, FG(Scorei) Represents the ratio of the good-label clients in the ith group to all clients in the entire ith group, FB(Scorei) Representing the ratio of the bad label client in the ith group to all clients in the whole ith group;
and determining the good-bad discrimination of the good-bad labels of the customers in the customer population matrix based on the KS value, thereby determining the offset degree of the customer population.
9. The method of claim 8, wherein scoring each customer in the customer population matrix according to the loan application scoring model comprises:
obtaining the performance of the customer group on a transverse time axis, and obtaining the score of the customer group based on the performance;
comparing the scores of the customers with the scores of the customer groups obtained by the loan application scoring model prediction to obtain a comparison result, and judging whether the comparison result meets the preset requirement or not;
if so, taking the score of the customer group obtained by the loan application scoring model prediction as the final score of the customer group;
otherwise, determining a comparison result and a deviation value required by a preset value, resetting a trigger threshold range of the loan application scoring model according to the deviation value, recording a corresponding model effect function value in the trigger threshold range every time, and selecting a trigger threshold with the maximum model effect function value as a new trigger threshold;
obtaining loan policy adjustment information, market competition change information and operation management adjustment information in a time period from the time when the credit granting application client is shifted to the time when the credit granting application client is shifted again, and determining policy influence factors, competition influence factors and management influence factors on the basis of the loan policy adjustment information, the market competition change information and the change degree of the operation management adjustment information;
extracting loan data related to the policy influence factor, the competition influence factor and the management influence factor from a loan database respectively;
obtaining an input data set based on the loan data, inputting the input data in the input data set into the loan application scoring model, outputting a prediction result, comparing the prediction result with a scoring result corresponding to the credit application client, and determining a main influence factor;
adjusting the risk strategy of the loan application scoring model by referring to other influence factors based on the main influence factor;
and determining the final score of the customer group based on the adjusted loan application scoring model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110983469.1A CN113919932A (en) | 2021-08-25 | 2021-08-25 | Client scoring deviation detection method based on loan application scoring model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110983469.1A CN113919932A (en) | 2021-08-25 | 2021-08-25 | Client scoring deviation detection method based on loan application scoring model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113919932A true CN113919932A (en) | 2022-01-11 |
Family
ID=79233255
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110983469.1A Pending CN113919932A (en) | 2021-08-25 | 2021-08-25 | Client scoring deviation detection method based on loan application scoring model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113919932A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117408805A (en) * | 2023-12-15 | 2024-01-16 | 杭银消费金融股份有限公司 | Credit wind control method and system based on stability modeling |
| CN117575769A (en) * | 2023-11-02 | 2024-02-20 | 睿智合创(北京)科技有限公司 | Credit agency customer flow quality assessment method and system |
-
2021
- 2021-08-25 CN CN202110983469.1A patent/CN113919932A/en active Pending
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117575769A (en) * | 2023-11-02 | 2024-02-20 | 睿智合创(北京)科技有限公司 | Credit agency customer flow quality assessment method and system |
| CN117408805A (en) * | 2023-12-15 | 2024-01-16 | 杭银消费金融股份有限公司 | Credit wind control method and system based on stability modeling |
| CN117408805B (en) * | 2023-12-15 | 2024-03-22 | 杭银消费金融股份有限公司 | Credit wind control method and system based on stability modeling |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN116342259A (en) | Automatic user credit rating method and device, electronic equipment and medium | |
| CN106980988A (en) | Trade company's value assessment method | |
| CN106709513A (en) | Supervised machine learning-based security financing account identification method | |
| CN106067088A (en) | E-bank accesses detection method and the device of behavior | |
| CN109377339B (en) | Auxiliary screening method and system for suspicious transaction cases | |
| CN112102073A (en) | Credit risk control method and system, electronic device and readable storage medium | |
| CN101194286A (en) | Risk based data assessment | |
| CN110909963A (en) | Credit scoring card model training method and taxpayer abnormal risk assessment method | |
| CN113919932A (en) | Client scoring deviation detection method based on loan application scoring model | |
| CN103366123A (en) | Software risk assessment method based on defect analysis | |
| CN107194815B (en) | Client segmentation method and system | |
| CN104850868A (en) | Customer segmentation method based on k-means and neural network cluster | |
| CN111062597A (en) | Method and device for detecting criminal suspicion of financial statement of listed company | |
| CN115543973A (en) | Data quality rule recommendation method based on knowledge spectrogram and machine learning | |
| CN112016858A (en) | Subjective and objective weighting method-based public opinion risk evaluation method | |
| Derks et al. | The Bayesian approach to audit evidence: Quantifying statistical evidence using the Bayes factor | |
| CN113919937B (en) | KS monitoring system based on loan assessment wind control | |
| CN115587887A (en) | Supply chain financial service-based team fraud identification method and device | |
| CN114418018A (en) | Model performance evaluation method, device, equipment and storage medium | |
| KR102499182B1 (en) | Loan regular auditing system using artificia intellicence | |
| CN111858639B (en) | External data management system and method for wind control management | |
| CN114626940A (en) | Data analysis method and device and electronic equipment | |
| CN114625781A (en) | Commodity housing value-based batch evaluation method | |
| CN113537759A (en) | User experience measurement model based on weight self-adaptation | |
| CN111612602A (en) | Suspected financial risk distinguishing method and device for listed company |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20220111 |