Credit examination result estimation system

Abstract

To accurately and easily perform credit examination.SOLUTION: A credit examination result estimate system comprises: acquisition means which acquires examination application data of an applicant in lease contract; storage means which stores a first learning model preliminarily generated by AI and machine learning which considers data in the application for the examination of the applicant in past credit examination in the lease contract, and a credit examination result for the applicant as teacher data; and estimate means which estimates whether a first credit examination result for the applicant is approval or rejection, on the basis of the examination application data of the applicant acquired by the acquisition means, by using the first learning model.SELECTED DRAWING: Figure 4

Description

[Technical field]

[0001] The present invention relates to a credit examination result estimation system. [Background technology]

[0002] In recent years, a scoring system has been proposed that quantifies the social credibility of a person or company as a score. For example, Patent Document 1 discloses a system that calculates a score indicating the credibility of a credit applicant according to credit parameters such as the age, place of employment, and annual income of the applicant, and a score calculation formula. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent No. 6561360 Summary of the Invention [Problem to be solved by the invention]

[0004] However, in order to improve the accuracy of credit decisions, there was a need to realize a system that would perform comprehensive credit screening based on application know-how.

[0005] The present invention has been made in consideration of the above circumstances, and has an object to provide a technology for easily conducting credit examinations with high accuracy. [Means for solving the problem]

[0006] In order to achieve the above-mentioned objective, one aspect of the present invention is a credit screening result estimation system having an acquisition means for acquiring application data for an applicant under a lease contract, a storage means for storing a first learning model generated in advance by AI and machine learning using data from the applicant at the time of application for screening in past credit screening under a lease contract and the credit screening results for the applicant as training data, and an estimation means for using the first learning model to estimate whether the first credit screening result for the applicant is approval or rejection based on the application data for the applicant acquired by the acquisition means. Effect of the Invention

[0007] According to the present invention, credit examination can be performed easily and with high accuracy. [Brief description of the drawings]

[0008] [Figure 1] FIG. 13 is a diagram showing an overview of credit result estimation. [Diagram 2] FIG. 1 is a diagram illustrating an outline of a system configuration. [Diagram 3] FIG. 2 is a block diagram showing the hardware configuration of a learning device and an estimation device. [Figure 4] FIG. 2 is a functional block diagram showing an example of a functional configuration of a learning device and an estimation device. [Diagram 5] 13 is a flowchart showing an example of a credit result estimation process. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0009] (Embodiment) <Summary> Hereinafter, this embodiment will be described with reference to the drawings. Fig. 1 is a diagram showing an example of credit result estimation to which an estimation device (credit result estimation device, information processing device) according to this embodiment is applied. In this embodiment, an example of performing credit examination in two stages, normal examination and risk control, will be described. The normal screening is a screening process that imitates the credit screening that has been conducted by screening personnel in the past. Risk control is a process to extract non-good customers that have been overlooked in the past and good customers that have been missed based on the results of the normal screening.

[0010] In the example of FIG. 1, when the credit application data CD of the credit applicant for a lease contract or the like is input, rejection / approval screening CS is performed on a rule basis. In the rejection / approval screening CS, a rule-based rejection or approval decision is made by referring to the past delinquency history, good customer information, etc. In the screening, if a rejection decision based on the delinquency history or an approval decision based on the good customer information is not made, the next screening process is performed. In the screening process, an approval / rejection decision CE1 is made, and as a result of the decision, it is determined whether the decision is approved CE2 or rejected CE3. In the screening process, a score (for example, 0 to 100) is output together with the result of the decision. At this time, a decision reason presentation CE4 is performed. As the reason for the decision, for example, the parameters that contributed to the decision and the contribution degree of the parameters are output. In the present embodiment, the approval / rejection decision CE1 is made based on a classifier (first learning model) generated by AI / machine learning and the review application data CD.

[0011] Furthermore, in the risk control process, if the above-mentioned screening process determines that the customer is approved CE2, a non-good customer judgment RC1 is performed, and the result of the judgment is determined to be either approval RC3 or rejection RC4. In the non-good customer judgment RC1, if the customer is approved in the screening process, non-good customers who have been overlooked up until now are added to the credit application data and judged based on performance data after past contracts. In this embodiment, the non-good customer judgment RC1 is judged based on the screening application data CD using a classifier (second learning model) generated by AI and machine learning.

[0012] In the risk control process, if the screening process determines that the application is rejected CE3, a good customer judgment RC2 is performed, and the result of the judgment is whether the application is approved RC3 or rejected RC4. In the good customer judgment RC2, if the application is rejected in the screening process, good customers who have been overlooked up until now are added to the credit application data, and a judgment is made based on past performance data after the contract is signed. In this embodiment, the good customer judgment RC2 uses a classifier (third learning model) generated by AI and machine learning to make a judgment based on the screening application data CD.

[0013] In the risk control process, whether the result is approval RC3 or rejection RC4 is determined according to the score output in the non-good customer judgment RC1 or good customer judgment RC2 and a predetermined threshold Th1 (e.g., 70). If the score is close to the predetermined threshold Th1, it is extracted by the gray zone extraction GE, and the judgment of the credit examination is left to the judgment AJ of the examiner as before. The gray zone extraction GE can also be considered as an alert to the examiner when the judgment of the credit examination cannot be made automatically. The result of the judgment by the examiner, whether approval or rejection, is output as the result of the risk control process. In the risk control process, a score (e.g., 0 to 100) is output together with the result of the judgment. In addition, in this case, the judgment reason presentation RC5 is performed in the same manner as the above-mentioned judgment reason presentation CE4.

[0014] When the risk control process outputs whether the application is approved or rejected, an examination response ER is created and output. The examination response ER indicates information about the credit applicant, the credit results in the credit process or risk control process, the score, and the reason for the judgment, etc. The examiner refers to the examination response ER to determine whether or not to enter into a contract CO with the credit applicant. The credit examination estimation system in this embodiment will be described in detail below.

[0015] <System configuration> 2 is a diagram showing an outline of the system configuration of the credit examination result estimation system according to this embodiment. The credit examination result estimation system according to this embodiment is configured by connecting a learning device 1 that learns a learning model used in estimating a credit result, a model DB 2 that stores the learning model, and an estimation device 3 that estimates a credit result (credit result estimation device) to each other via a predetermined network N such as the Internet.

[0016] <Hardware configuration> 3 is a block diagram showing the hardware configuration of the learning device 1 according to this embodiment. The learning device 1 includes a CPU (Central Processing Unit) 11, a ROM (Read Only Memory) 12, a RAM (Random Access Memory) 13, a bus 14, an input / output interface 15, an output unit 16, an input unit 17, a storage unit 18, a communication unit 19, and a drive 20.

[0017] The CPU 11 executes various processes according to a program recorded in the ROM 12 or a program loaded from the storage unit 18 to the RAM 13. The RAM 13 also stores data and the like required for the CPU 11 to execute various processes. The CPU 11, the ROM 12, and the RAM 13 are interconnected via a bus 14. An input / output interface 15 is also connected to this bus 14.

[0018] An output unit 16, an input unit 17, a memory unit 18, a communication unit 19, and a drive 20 are connected to the input / output interface 15. The output unit 16 is composed of a display, a speaker, etc., and outputs various information as images and sounds. The input unit 17 is composed of a keyboard, a mouse, etc., and inputs various information. The memory unit 18 is composed of a hard disk, a DRAM (Dynamic Random Access Memory), etc., and stores various data. The communication unit 19 communicates with other devices via a network N including the Internet.

[0019] A removable medium 21 consisting of a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, or the like is appropriately attached to the drive 20. A program read from the removable medium 21 by the drive 20 is installed in the storage unit 18 as necessary. The removable medium 21 can also store various data stored in the storage unit 18 in the same manner as the storage unit 18. Although not shown, the estimation device 3 has a hardware configuration shown in FIG. 3.

[0020] <Functional configuration> FIG. 4 is a functional block diagram showing an example of the functional configuration of the learning device 1, the model DB 2, and the estimation device 3 according to this embodiment.

[0021] <Learning device 1> In the CPU 11 of the learning device 1, an acquisition unit 111, a preprocessing unit 112, a learning unit 113, a relearning unit 114, and the like function during operation.

[0022] The acquisition unit 111 acquires training data 501 for AI and machine learning. The acquired training data is recorded, for example, in the storage unit 18. The training data 501 includes data for performing training in each of a first learning model used in the approval / rejection decision CE1 of the screening process shown in FIG. 1, a second learning model used in the non-good customer decision RC1, and a third learning model used in the good customer decision RC2.

[0023] For example, the training data for learning the first learning model includes data of the applicant at the time of application for screening in past credit screening for rental contracts and the credit screening results for the applicant. Examples of the data at the time of application for screening include the applicant's nationality, sex, rent, employment type, annual income, date of birth, etc., and open data (external open data that is technically difficult to use in general) regarding the region and location (related to the subject of the rental contract). In this embodiment, the credit screening results include either approval or rejection.

[0024] For example, the training data for the second learning model includes, in addition to the data at the time of application for screening and open data, the applicant's mid- to long-term delinquency record data after the contract, bad debt and litigation eviction record data (history of tenants after the contract regarding mid-term delinquency, bad debt, and litigation eviction), and the credit screening results for the applicant. Mid- to long-term delinquency record data is, for example, data regarding delinquency in rent, etc. after the applicant enters into a rental contract, and particularly data regarding the record of mid- to long-term delinquency in rent, etc. The bad debt and litigation eviction record data is, for example, data regarding the presence or absence of bad debts in rent, etc., after the applicant enters into a rental contract, as well as the number of times, the period, and the record of eviction due to litigation.

[0025] Furthermore, for example, the teacher data for learning the third learning model includes, in addition to the data at the time of application for the screening and the open data, the applicant's past record of delinquency after the contract, past record of short-term repayment (history of the tenant after the contract), and the credit screening results for the applicant. The past record of delinquency is, for example, data on whether or not the applicant has delinquent payments of rent, etc. after signing a rental contract. The past record of short-term repayment is, for example, data on whether or not the applicant has delinquent payments of rent, etc. after signing a rental contract, paid off in full in the short term (within a specified period of time).

[0026] The preprocessing unit 112 performs various preprocessing operations on the training data for learning. For example, one example of the preprocessing operation is a process (data augmentation) in which various processes are performed on the original data (training data) to increase the number of training data. Note that the preprocessing operation does not necessarily have to be performed.

[0027] In this embodiment, the learning unit 113 performs learning processing for each of the first learning model, the second learning model, and the third learning model using the above-mentioned teacher data. The first learning model, the second learning model, and the third learning model are estimation engines that estimate the result of credit screening by regression based on the data at the time of the screening application, etc. The first learning model, the second learning model, and the third learning model can be learned by any supervised learning algorithm. In this embodiment, ensemble learning that combines gradient boosting (GBM) and deep neural network, etc., is adopted as the learning algorithm, but various other learning algorithms such as support vector machine (SVM), light GBM, and neural network can be adopted. In ensemble learning that combines multiple learning algorithms to create one learning model, the optimal model is automatically used.

[0028] As a result of the learning process by the learning unit 113, a first learning model, a second learning model, and a third learning model are generated that accept the data at the time of the above-mentioned examination application as input and estimate the result of the credit examination for the applicant. The first learning model, the second learning model, and the third learning model may be configured to output the reliability (score) of the credit examination together with the result of the credit examination.

[0029] The re-learning unit 114 re-learns the first learning model at a predetermined period (e.g., monthly) using the data at the time of the above-mentioned screening application and the screening results output as a result of the risk control shown in FIG. 1 as teacher data. This makes it possible to improve the accuracy of the output (screening results) in the learning model (first learning model) in accordance with trends in the world. Note that the screening results (feedback of the judgment results) output as a result of the risk control are acquired by the acquisition unit 111. The re-learning unit 114 can also be considered as re-learning the first learning model based on a collection of learning data including the applicant's screening application data acquired by the acquisition unit 111 and teacher data which is the final credit screening result for the applicant, and storing the model in the model DB2.

[0030] <Model DB2> The model DB 2 includes a learning model DB 201 that stores the above-mentioned learning models. The learning models may be stored in a storage unit in the learning device 1 or the estimation device 3.

[0031] ≪Estimation device 3≫ In the CPU 31 of the estimation device 3, an acquisition unit 311, a preprocessing unit 312, an estimation unit 313, a determination reason presentation unit 314, a gray zone extraction unit 315 (alert means), an output unit 316, and the like function when operating.

[0032] The acquisition unit 311 acquires the applicant's examination application data for the rental contract (credit examination data, data at the time of examination application). The acquired credit examination data is stored in a storage unit (not shown) in the estimating device 3.

[0033] The pre-processing unit 312 performs pre-processing on the applicant's examination application data acquired by the acquisition unit 311. For example, the pre-processing unit 312 performs rejection screening on a rule-based basis based on data on the applicant's past delinquency history. Also, for example, the pre-processing unit 312 performs approval screening on a rule-based basis based on a good customer list.

[0034] The estimation unit 313 uses the first learning model of the model DB2 to estimate whether the credit examination result for the applicant (hereinafter referred to as the first credit examination result) is approval or rejection based on the applicant's examination application data acquired by the acquisition unit 311. In addition, the estimation unit 313 uses the second learning model of the model DB2 to estimate the final credit examination result for the applicant (hereinafter referred to as the second credit examination result) based on the examination application data of the applicant whose first credit examination result is estimated to be approval. In addition, the estimation unit 313 uses the third learning model of the model DB2 to estimate the final second credit examination result for the applicant based on the examination application data of the applicant whose first credit examination result is estimated to be rejection. The estimation unit 313 outputs the credit examination result, the reliability, and the parameters (e.g., annual income) that contributed to the judgment in the above-mentioned credit examination, and the influence of the parameters (contribution degree; for example, 40%).

[0035] The reason for judgment presentation unit 314 outputs the reason for judgment including the parameters that contributed to the judgment and the degree of influence for each credit examination result of the estimation unit 313. In the credit examination, if the contributions of annual income and employment type are 20% and 40%, respectively, the fact is output (presented) as the reason for judgment.

[0036] The gray zone extraction unit 315 outputs an alert when the reliability of the screening result output by the second learning model or the third learning model is within a predetermined range. For example, the predetermined range may be a range of several percent around a predetermined threshold (for example, if the predetermined threshold is 70, the score ranges from 69 to 71). The above-mentioned alert is output, for example, to a terminal (not shown) of a person in charge of credit screening, and the subsequent credit screening is performed by the person in charge. Then, the estimation unit 313 may set the credit screening result determined by the person in charge in response to the above-mentioned alert as the final second credit screening result.

[0037] The output unit 316 outputs the final second credit examination result output by the estimation unit 313 to the above-mentioned learning device 1. The learning device 1 performs the above-mentioned re-learning based on the output data.

[0038] <Processing content> FIG. 5 is a diagram showing an example of the credit result estimation process according to the present embodiment. In step S1, the acquisition unit 311 acquires the examination application data. In step S2, the pre-processing unit 312 performs the above-mentioned rejection / approval screening based on the application data. This allows for a rule-based rejection (or approval) decision based on specific conditions (such as so-called blacklists or specific good customers) according to past delinquency history, etc., so that a rejection decision can be made quickly.

[0039] In step S3, the estimation unit 313 uses the first learning model to make an approval / rejection decision in the above-mentioned credit screening (obtains the output result of the first learning model). The screening process in step S3 can be considered as an automation of screening that was previously performed by credit officers. In other words, the first learning model, which has been subjected to AI and machine learning based on know-how accumulated by credit officers over many years, outputs a result close to that of a screening by a credit officer. Also, in step S3, the judgment reason presentation unit 314 outputs the parameters that contributed to the judgment and the reason for the judgment, including the degree of influence, in the credit screening result.

[0040] In step S4, the estimation unit 313 judges whether the result of the credit examination in step S3 is approval or denial. If it is approval, the process proceeds to step S5, and if it is denial, the process proceeds to step S6.

[0041] In step S5, the estimation unit 313 uses the second learning model to make an approval / rejection decision in the above-mentioned credit screening (obtains the output result of the second learning model). The risk control process in step S5 can be considered as determining non-good customers who have been overlooked in the past among the applicants who have been determined to be approved in step S3. In other words, customers who have a high score based on the screening application data but are at risk of mid- to long-term delinquency, bad debts, or eviction due to litigation are extracted based on past performance data.

[0042] In step S6, the estimation unit 313 uses the third learning model to make an approval / rejection decision in the above-mentioned credit screening (obtains the output result of the third learning model). The risk control process in step S6 can also be considered as determining good customers who were previously overlooked among the applicants who were rejected in step S3. In other words, customers who have a low score based on the screening application data but are expected to not be in arrears or to pay off their debts in a short period of time based on past performance data are extracted.

[0043] In step S7, the gray zone extraction unit 315 judges whether the reliability of the examination result in step S5 or S6 is within a predetermined range. If the reliability is within the predetermined range, the process proceeds to step S8. If not, the process proceeds to step S10. In step S8, if the reliability of the examination result in step S5 or S6 is within a predetermined range, the gray zone extraction unit 315 outputs an alert to the terminal of the credit officer. In step S9, the acquisition unit 311 acquires the result of the credit examination determined by the person in charge in response to the above-mentioned alert.

[0044] In step S10, the output unit 316 outputs the final second credit examination result output by the estimation unit 313 to the above-mentioned learning device 1. Note that, when a person in charge makes a judgment in response to the above-mentioned alert, the estimation unit 313 sets the credit examination result of the person in charge as the final second credit examination result. Also, in step S10, the judgment reason presentation unit 314 outputs the parameters that contributed to the judgment and the judgment reason including the degree of influence in the credit examination result.

[0045] <Advantageous Effects of the Present Embodiment> According to the above-described embodiment, by using a learning model that has been trained based on large amounts of data and know-how (credit screening results) obtained during the operation of past credit screening operations, it is possible to easily and quickly perform highly accurate credit screening. Furthermore, according to the above-described embodiment, as a risk control process, a learning model trained based on performance data after past contracts is used, making it possible to extract non-good customers that have been previously overlooked and good customers that have previously been missed.

[0046] Although one embodiment of the present invention has been described above, the present invention is not limited to the above-described embodiment, and modifications, improvements, etc. within the scope that can achieve the object of the present invention are included in the present invention.

[0047] (others) Also, for example, the above-mentioned series of processes can be executed by hardware or software. In other words, the above-mentioned functional configuration is merely an example and is not particularly limited. That is, it is sufficient that the information processing system is provided with a function capable of executing the above-mentioned series of processes as a whole, and the type of functional block used to realize this function is not particularly limited to the above example. In addition, the location of the functional block is not particularly limited to that shown in FIG. 4 and may be arbitrary. For example, the functional block of the estimation device (server) may be transferred to another device or the like. Conversely, the functional block of the other device may be transferred to the estimation device or the like. In addition, one functional block may be configured as a single piece of hardware, may be configured as a single piece of software, or may be configured as a combination thereof.

[0048] When a series of processes is executed by software, a program constituting the software is installed in a computer or the like from a network or a recording medium. The computer may be a computer built into dedicated hardware. The computer may also be a computer capable of executing various functions by installing various programs, such as a server, a general-purpose smartphone, or a personal computer.

[0049] The recording medium containing such a program may be configured not only as a removable medium (not shown) distributed separately from the device main body in order to provide the program to the user, etc., but also as a recording medium etc. provided to the user in a state in which the medium is pre-installed in the device main body. Since the program can be distributed via a network, the recording medium may be mounted on or accessible to a computer connected or connectable to the network.

[0050] In this specification, the steps describing the program recorded on the recording medium include not only processes performed in chronological order, but also processes that are not necessarily performed in chronological order but are performed in parallel or individually. In addition, in this specification, the term "system" refers to an overall device composed of multiple devices, multiple means, etc.

[0051] In other words, the credit examination result estimation system to which the present invention is applied can take various forms having the following configurations. That is, the credit examination result estimation system includes an acquisition means (e.g., acquisition unit 311) for acquiring the applicant's examination application data in the lease contract, a storage means (e.g., learning model DB 201) for storing a first learning model generated in advance by AI and machine learning using the applicant's examination application data in past credit examinations in the lease contract and the credit examination results for the applicant as teacher data, and an estimation means (e.g., estimation unit 313) for estimating whether the first credit examination result for the applicant is approval or rejection based on the applicant's examination application data acquired by the acquisition means using the first learning model. As a result, the first learning model, which has been subjected to AI and machine learning based on the know-how accumulated by the credit officer for many years, can output a result close to that of the examination by the credit officer.

[0052] The storage means may store a second learning model generated in advance by AI / machine learning using training data including data on the applicant at the time of application for screening in past credit screenings for lease contracts, mid- to long-term delinquent records after the contract for the applicant, bad debt / lawsuit eviction records, and the credit screening results for the applicant, and the estimation means may use the second learning model to estimate the final second credit screening results for the applicant based on the screening application data for the applicant whose first credit screening result was estimated to be approved by the estimation means. This makes it possible to extract non-good customers who have been overlooked in the past.

[0053] The system may further include an alert means (e.g., gray zone extraction unit 315) that outputs an alert when the reliability of the screening result output by the second learning model is within a predetermined range, and the estimation means may determine the credit screening result determined by the person in charge in response to the alert as the final second credit screening result. This allows the credit screening person to make the decision when the score output as the final second credit screening result is close to a threshold (when it is difficult to determine by automation).

[0054] The second learning model may be a learning model generated in advance by AI and machine learning using training data including data on the applicant's application for screening in past credit screenings for the lease contract, mid- to long-term delinquent records after the contract for the applicant, bad debt and litigation eviction records, and external open data that is technically difficult to use in general, as well as the results of the credit screening for the applicant. This makes it possible to output results closer to screening by credit officers using the second learning model that has been subjected to AI and machine learning based on training data closer to the know-how accumulated over many years by credit officers.

[0055] The storage means may store a third learning model generated in advance by AI / machine learning using training data including data on the applicant's application for screening in past credit screenings for lease contracts, the applicant's past delinquent payment data after the contract, and the short-term full repayment data, as well as the credit screening results for the applicant, and the estimation means may use the third learning model to estimate the final second credit screening results for the applicant based on the screening application data for the applicant whose first credit screening result was estimated to be rejected by the estimation means. This makes it possible to extract good customers who have been overlooked in the past.

[0056] In addition, the system may further include an alert means for outputting an alert when the reliability of the examination result output by the third learning model is within a predetermined range, and the estimation means may determine the credit examination result determined by the person in charge in response to the alert as the final second credit examination result. The credit examination result estimation system according to claim 5. This makes it possible to leave the decision to the credit officer when the score output as the final second credit examination result is close to the threshold (when it is difficult to determine by automation).

[0057] The third learning model may be a learning model generated in advance by AI and machine learning using training data including data on the applicant's application for screening in past credit screenings for the lease contract, the applicant's past delinquent payment data after the contract, short-term full repayment data, and external open data that is technically difficult to use in general, as well as the results of the credit screening for the applicant. This makes it possible to output results closer to screening by credit officers using the third learning model that has been subjected to AI and machine learning based on training data closer to the know-how accumulated by credit officers over many years.

[0058] It is also preferable to have a means for presenting judgment reasons (e.g., judgment reason presenting unit 314) for outputting judgment reasons including parameters that contributed to the judgment and their degrees of influence in the judgment of the estimation means. This makes it possible to show how much each parameter influenced the result in an automatically performed credit examination. This also makes it possible to verify whether the parameters and their degrees of contribution are close to those that a previous credit officer contributed to his or her own judgment.

[0059] The system may further include a re-learning means (e.g., a re-learning unit 114) for re-learning the first learning model based on a set of learning data including the applicant's screening application data acquired by the acquisition means and teacher data which is the final credit screening result for the applicant, and storing the model in the storage means. This makes it possible to improve the accuracy of the output (screening result) of the learning model (first learning model) in accordance with world trends. [Explanation of symbols]

[0060] 3: Estimation device 31: CPU 32: Communication department 201: First learning model 311: Acquisition unit 313: Estimation unit

Claims

1. An acquisition means for acquiring application data of an applicant for a lease contract; A storage means for storing a first learning model that has been generated in advance by AI / machine learning using training data of the applicant's application for credit screening in past credit screenings for lease contracts and the results of credit screenings for the applicant; and an estimation means for estimating, using the first learning model, whether a first credit examination result for the applicant is approval or rejection based on the applicant's application data acquired by the acquisition means, The storage means stores a second learning model that has been generated in advance by AI / machine learning using training data including data on the applicant at the time of application for screening in past credit screening for lease contracts, mid- to long-term delinquent record data after the contract for the applicant, bad debt / lawsuit eviction record data, and the credit screening results for the applicant; The estimation means estimates, using the second learning model, a final second credit examination result for the applicant based on the application data for the applicant whose first credit examination result has been estimated to be approval by the estimation means. Credit review result estimation system.

2. The method further includes an alert means for outputting an alert when the reliability of the examination result output by the second learning model is within a predetermined range; The estimation means determines the result of the credit check determined by the person in charge in response to the alert as the final second credit check result. The credit examination result estimation system according to claim 1 .

3. The storage means stores a third learning model that has been generated in advance by AI / machine learning using training data including data on the applicant at the time of application for screening in past credit screening for rental contracts, data on past delinquency records after the contract for the applicant, data on short-term full repayment records for the applicant, and the results of credit screening for the applicant; The estimation means estimates, using the third learning model, a final second credit examination result for the applicant based on the application data for the applicant whose first credit examination result has been estimated to be rejected by the estimation means.

3. A credit examination result estimation system according to claim 1 or 2.

4. Further, an alert means is provided for outputting an alert when the reliability of the examination result output by the third learning model is within a predetermined range; The estimation means determines the result of the credit check determined by the person in charge in response to the alert as the final second credit check result. The credit examination result estimation system according to claim 3.

5. The method further includes a judgment reason presentation means for outputting a judgment reason including a parameter that contributed to the judgment and an influence degree in the judgment of the estimation means. The credit examination result estimation system according to any one of claims 1 to 4.

6. and a re-learning means for re-learning the first learning model based on a set of learning data including the applicant's application data acquired by the acquiring means and teacher data which is a final credit screening result for the applicant, and storing the re-learned model in the storage means. The credit examination result estimation system according to any one of claims 1 to 5.

Images