WO2019021314A1

WO2019021314A1 - A system and method for default probability prediction and credit scoring framework

Info

Publication number: WO2019021314A1
Application number: PCT/IN2018/050488
Authority: WO
Inventors: Aviruk CHAKRABORTY
Original assignee: Capitaworld Platform Private Limited
Priority date: 2017-07-26
Filing date: 2018-07-25
Publication date: 2019-01-31
Also published as: WO2019021314A4

Abstract

The present invention discloses a system and method for default probability prediction and credit scoring framework. The present invention comprises various components such as fetching means includes a video source (10) performing recognition process like, face recognition (4a), an emotion detection (4b), pulse detection (4c), speech analysis (4d), galvanic skin response (4e); a host or central computing unit (11), computer accelerators (13) and a storage updated database (14b) to collect user data. The said insights analysis is processed further with all relevant financial as well as social media information. The applicant's application with all said parameters is combined and sent for auto filing without human intervention via a deep neural network model (14a). This process is centrally focused with autonomous process predicts default and credit score of the applicant and eases out the work for any person seeking a loan and other side as well for who is providing the loan or insurance.

Description

A SYSTEM AND METHOD FOR DEFAULT PROBABILITY PREDICTION AND CREDIT SCORING FRAMEWORK

The present invention is a combination of Provisional Patent Application No. 201721026520 Filed on July 26, 2017 and Provisional Patent Application No. 201721026525 Filed on July 26, 2017.

Field of the invention The present invention relates to a system of default probability prediction and credit scoring framework. More particularly it relates to a method to diagnose credit scoring and default probability prediction using face detection and other touchless diagnostics cognitive like emotion quantification, financial information as well as digital footprint which facilitates the loan or lending services with completely digitalized approach using the deep neural network and smoothens out the entire loan processes.

Background of the invention

The prediction of loan defaults has been the basis of a growing interest in the development of systems of credit scoring. Typically, discriminant analysis, logit or some other type of classificatory procedure has then been applied to develop a model for distinguishing between "good" and "bad" payers. Credit scoring models are usually formulated by fitting the probability of loan default as a function of individual evaluation attributes. Typically, these attributes are measured using a likert-type scale, but are treated as interval scale explanatory variables to predict loan defaults. Existing models also do not distinguish between types of default, although they vary: default by an insolvent company and default by an insolvent debtor.

Individual borrowers pay their loans or loan installments when they have the ability to pay. The ability to pay largely depends on a person's disposable income. And if a person's disposable income disappears due to the loss of his job, or due to income reduction resulting from a pay cut or a change in job or due to underemployment, then the person assumes a much higher risk of defaulting on his loan repayments simply because the person has no money and therefore has no ability to pay. That is why it is critical to predict a person's ability to pay based on his future probability of loss of income or a reduction in income in order to make a superior prediction of his creditworthiness. Today, the standard approach to credit scoring is through traditional credit scores but the problem is that they are increasingly becoming inaccurate, simply because they don't predict future ability to pay. They are essentially reactive scores, meaning they change after borrowers default, and do not factor changes in the economy, and purely rely on credit histories and consumers' past ability to pay.

Conventionally a system was pioneered as a credit scoring method that has become widely accepted by lenders as a reliable means of credit evaluation, helping determine the likelihood that how credit users (i.e. borrowers) will pay on their debts. A score is generated from this credit scoring method which condenses a borrower's credit history into a single number. Credit scores are calculated by using scoring models and mathematical tables that assign points for different pieces of information which approximate a borrower's future credit performance. Developers of the score-model find predictive factors in the data that can indicate future credit performance. For instance, predictive factors such as the amount of credit used versus the amount of credit available, length of time at a present employer, and negative credit information such as bankruptcy can be revealed in a borrower's credit history. The problem is that in many parts of the world, collectively known as the emerging markets a borrower's credit history cannot be determined because the lending infrastructure does not exist.

Probability of default (PD) is a financial term describing the likelihood of a default over a particular time horizon. It provides an estimate of the likelihood that a borrower will be unable to meet its debt obligations. Probability of default is used in a variety of credit analyses and risk management frameworks. It is a key parameter used in the calculation of economic capital or regulatory capital for a banking institution. Probability of default is the risk that the borrower will be unable or unwilling to repay its debt in full or on time. The risk of default is derived by analyzing the obligator's capacity to repay the debt in accordance with contractual terms. Probability of default is generally associated with financial characteristics such as inadequate cash flow to service debt, declining revenues or operating margins, high leverage, declining or marginal liquidity, and the inability to successfully implement a business plan. In addition to these quantifiable factors, the borrower's willingness to repay also must be evaluated.

The traditional or conventional method of loan process is very time consuming and having loads of paperwork. Moreover, it also requires the person who is claiming for loan to visit the bank or contact the banker. This is not possible always for any person to spare this much time. Currently credit scoring is decided based on a person's financial history and transactions. They are all checked against the timeline if the installments were paid on time or not. This is just one aspect and there are some more which are sales, expenditure, profit with and without tax, operating profit, liabilities and equities, shareholders' details, assets and margins. The risk is calculated based on the information provided by the loan seeker while applying for the loan. There are central agency and international agencies which maintain such data for any user. They also get the background check done in case of new loan seeker who does not have a history of financial security background or pending/on-going loans. When there is no data, the decision becomes a little difficult. Moreover, people may get the fake documents or history modified with extra paperwork which are in fact important for the decision making and rating that person' s score.

A practice to predict default probability (failure to meet legal obligations/conditions of a loan) is based on the history and documents. All the past financial records starting with credit scoring, industry performance, market references and other histories of on-going loans are taken into consideration. Depending upon the country the weightage for the parameters may differ. The credit scoring does consider that each installment is paid in time or within certain duration; the history is being maintained for future reference. If there are multiple cases of default in the same industry, that industry can be under a category with more security requirements. Similarly, market references may also throw lights on the loan seeker to check if there was any problematic behavior in the past in order to stay alert. One of the other things may be the timely installment payments of other on-going loans. In case, one of them has a delayed payment, there might be a case in the near future when the current loan can also be affected and get delayed payment. However, if there is any good or bad history of any of such cases for the person's transaction or loans then it becomes difficult to predict future actions about loan repayment. In the world of Finance, where works have been performed by traditional approach, if a person isn't having any financial knowledge or awareness; he or she may enter a risk of impending with huge financial loss or lack of financial guidance which is a huge loss in all. Hence, to eradicate and reduce this financial risk with the help of wealth management, an improving credit scoring system based decision is extremely needed which is highly defined by the distinctively designed completely automated process. The current conventional approach uses various patterns for the person who seeks loan/investment. The primary reasons why bank and other financial institutions do not consider small and micro enterprises to provide financial support is that they have no credit record or meaningful assets. As a consequence, these enterprises are unable to access credit through conventional banking channels as these channels need financial history, accurate personal data and a credit history of a borrower, enabling assessment of the credit risk. The entrepreneurs running such enterprises end up accessing credit through alternative channels and from illegal syndicates which provide financial support at the exorbitant rate of interest. Many of these enterprises are unable to payback their loan amount on time due to heavy interest rates and end up in shutting down their business. In general, loan defaults can result from bankruptcies, closures, delays by owners, and the owners' poor credit. Bankruptcy is the legal condition of financial failure, while the closure is a permanent end to the company's business. Delay occurs when the debtor asks for the postponed redemption of a debt until it reaches a better state of corporate management. Poor credit represents the state of delinquency for more than three months. Hence there is a need for a system and method for ascertaining a credit worthiness of a potential borrower in order to provide financial support for his/her business growth.

Various prior arts have been disclosed which describing about predicting consumer credit risk with various assessment method. The prior art document US8799150B2 disclosing a method, to evaluate an individual's creditworthiness using income risk based credit score thereby providing creditors, lenders, marketers, and companies with consumer's credit risk and repayment potential. Further the income risk based credit scoring model's databases are updated by using an updated assessment of economic conditions, allowing the most current information to be used by lenders, businesses, and others. Thus, it quantifies the source of credit risk, by the interaction between the economy and consumer's income prospects, for defaulting on a payment. While it does not quantify or predict the default probability based on minute observations providing deep insights into a future probability over the existing methods. Therefore existing credit scoring models fail to take into account the consumer's true "capacity" to pay or the ability to pay which depends on consumer's future continuance or income risk.

Another prior art document U.S. 6,513,018 describes a statistical strategy for generating a credit score predictive of the likelihood of the desired performance result over the same relevant time interval for a selected credit user. However, such conventional techniques often fail to address inter-correlation between various variables within the collected credit user information, especially at the time of generation and/or optimization of process models, to correlate certain credit user information to certain credit risks simultaneously.

Above mentioned prior arts do not apparently describe about default probability prediction, and also there is no clarity regarding credit scoring system applied with fully automated model especially for loan automation system and financial requirements. Thus, there is an increasing need in the marketplace to incorporate the ability to pay risk in credit scoring models in order to decrease credit lending risks. Recognizing this unmet need, the present invention provides a highly predictive model that predicts the ability to pay component. This results in a new consumer risk score.

Therefore, it would be highly desirable to have a completely digitalized neural network system, which provides centrally focused process which eases out the work for any person for seeking loan and for other side as well who is providing the loan or insurance, and also the system is needed which reduces documentation, complex paperwork and tiresome process. Today, in the age of the digital world as everyone is moving towards the completely digitally focused approach, one of the necessity each and every one encounter is a need of "Loan" be it a business loan, personal loan, home loan, car loan and pay-day loan. By hearing the word "Loan", any person get a bit nervous due to its long tiresome process as well as documentation required which makes any person panic. So, to eradicate this situation, an optimal solution of "Credit score" along with default probability prediction has been introduced in the present invention which works as a savior for each and every one.

Object of the invention

The main object of present invention is to provide a system and method for diagnosing credit scoring by default probability prediction from a reading of face.

Another object of present invention is to provide a system and method for default probability prediction with facial analysis parameters and credit score which facilitates the lending services with the completely digitalized approach.

Yet another object of present invention is to provide a system and method for default probability prediction and credit scoring framework which reduces documentation, complex paperwork and tiresome process and not only partially but fully automates loan process for the consumer as well as the bank.

Still another object of the present invention is to provide a system and method for default probability prediction and credit scoring framework which works upon the presence of social media and financial information search with real time data.

The further object of the present invention is to provide a system and method for default probability prediction and credit scoring framework with the help of fine tune parameters and creates a deep neural network model approach.

Yet, the further object of the present invention is to provide a default probability prediction and credit scoring framework system and method which helps for risk management, monitors and tracks the transactions and helps out in wealth management.

Summary of the Invention

The present invention discloses a system and method for default probability prediction and credit scoring framework by inputting various facial analysis parameters from the camera capturing video source. Further it facilitates said outcomes to get an insight of an applicant's future probability of default using deep neural network model and maintains the credit score.

The present invention comprises main system components by fetching means which includes a video source receiving the video in raw format; a host or central computing unit, computer accelerators and a storage database. Said the fetching means perform recognition process like, face recognition, an emotion detection, pulse detection, speech analysis and galvanic skin response through video source to collect user data. The fetching means are interfaced to processor based host or central computing unit. The computing unit is connected with computation accelerators for parallel computation. The user data fetched via fetching means is stored and secured in data storage with the help of host or central computing unit. Further, all the computing units are having shared common memory which has access to the already learnt parameters and helps for further analysis. Further, when customer or applicant enters in for a loan or appears digitally with a video conference/call or a video recording and applies for any specific loan, the video source captures specific frame which detects the face and commences analysis by all said fetching means. The analysis of an applicant with said parameters is processed further with relevant financial information as well as social media presence and sent for auto filing without human intervention via the smart model. The data lake has been created with the help of data transmission and creates a deep neural network approach evaluating real time data which is a completely automated process. After that, it tracks a similar profile/pattern of an applicant and predicts the default probability as well as credit score and makes an alert. This process is centrally focused with an autonomous process which eases out the work for any person for seeking a loan and for other side as well who is providing the loan or insurance. After loan disbursement, the system of the present invention also monitors, tracks, and evaluates the disbursed loan for enhancing the existing credit scoring system.

Brief Description of the Drawings Fig. 1 shows a method for the main operational steps of default probability prediction and credit scoring framework system. Fig. 2 shows a block diagram for system of default probability prediction and credit scoring framework according to present invention.

Detailed description of the Invention

The nature of the invention and the manner in which it works is clearly described in the complete specification. The invention has various elements and they are clearly described in the following pages of the complete specification. Before explaining the present invention, it is to be understood that the invention is not limited in its application.

Before explaining the present invention, it is to be understood that the term Fetching means (10) via video source hereinafter refers to face recognition (4a), Emotion detection (4b), pulse detection (4c), speech analysis (4d) and Galvanic skin response (4e).

In the said process of the present invention, the camera capturing video source (10) fetches specific data of a person applying for a loan via fetching means as well as by other evaluation attributed such as social presence and financial data. The said process inputs all the data of a user into smart model for prediction of loan defaults and provides a credit score.

The credit scoring and default probability prediction system of the present invention comprises the camera capturing video source (10) for face detection followed by recognition of particularly face (4a), emotion detection (4b), pulse detection (4c), speech analysis (4d) and galvanic skin response (4e). A host or central computing system (11) being adapted for heterogeneous computing and training the credit scoring calculation model. Further, the system of the present invention comprises computation accelerators (13) to take down heavy mathematical computation, which makes this probability prediction and credit scoring process faster and smoother. Said computer implemented system quantifies applicant's credit risk and default probability by analysing each frame using vision methods through camera capturing video source (10). In this present invention, as shown in fig. 2, the whole system is processor- based. The host or central computing system (11) is connected with the plurality of computation accelerators (13) used for parallel computation. Here, both the host or central computing system (11) and computation accelerators (13) shared a common memory (12) which has access to the already learnt parameters and aids in supplementary analysis. The system of the present invention uses the data transmission with this hardware and is used for forming or updating an internal storage (14b). With an application software of a deep neural network model (14a) the information filtering system pre-fetch the data from the common memory (12). This data is useful in training the model of the present invention. For training purpose said model uses the host or central computing system (11) as for taking out critical functional processes, while the computation accelerator (13) takes out parallel processing and computation. Memory management is carried out by the deep neural network model (14a). Finally, upon completion of training, the saved data is stored to user specific data in internal storage (14b). As shown in figures 1 and 2, in starting with the process, is to develop a computer implemented system for quantifying applicant's credit risk and default probability by analysing each frame using vision methods through camera capturing video source (10). At an input step (1), a customer or an applicant or any person who is in the need of loan would enter in the loan automation system of present invention and apply for a loan, the loan provider meets the applicant in person or via digitally with a video conference/call or a video recording where the face of a person is identified and fetched via fetching means (10). The loan can be any one of the following types such as a business loan, personal loan, home loan, car loan, education loan and pay-day loan.

At step (2) the fetching means (10) is interfaced to a processor based on a host or central computing system (11) pre-f etches the necessary relevant data or information of a person such as historical data, real-time data, etc, from the shared memory (12).

At step (3) the fetching means (10) comprising a video source for identification process like face recognition (4a), emotion detection 4(b), pulse detection (4c), speech analysis (4d) and galvanic skin response (4e).

Further, when the face is detected, the video frame recognizes the historical data of an applicant, and once it is confirmed advance analysis is started. At step (4) the first analysis is face recognition (4a) which checks if the applicant is the same person in the video by matching uploaded profile picture and person found in the video frame. The second analysis is based on emotion detection (4b), the face is tracked continuously and the minute details are measured in each frame. When an applicant is worried the forehead temples get in tension or raise the eyebrows in a situation of surprised or doubt. The few inferences are analysed via emotion detection (4b) which cannot be avoided whether it could be sudden encounters to questions related to stress or unexpected discussion with the applicant. The third analysis is based on pulse detection (4c) which locates the forehead. The forehead is place having a flat skin area and helps focus on the centre of it for monitoring. The captured frame varies based on the blood flow on the face and a proper illumination enhances the information captured through the color channels of the image. While monitoring the face it measures the average optical intensity in forehead. The optical absorption characteristics of (oxy-) haemoglobin provides the physiological data in the form of a pulse. In such situations if the applicant either fakes or gets nervous or shocked, it creates major fluctuations or sudden fall/jump in the blood flow and provides an insight into the sudden change in applicant's answer pattern. The fourth analysis for prediction of default probability and credit score is speech analysis (4d). Through the fetching means via video source (10) the audio is recorded from the discussion between the loan provider and applicant, the audio fetched is used to analyze the tone and pitch of an applicant and emotions are recognized from voice generated at step (5) via voice analysis (5a). Further after the speech is analysed, it is converted to text format, which provides a sentiment analysis (5b) based upon the audio fetched via video source (10). The fifth analysis includes physiological data which is galvanic skin response (4e). It is monitored through the conversation via video source (10), the applicant's skin temperature, respiration, perspiration and other bodily evaluation is identified. At input step (6) the loan automation system of present invention scrapes the Internet/social media (6a) and all relevant databases for user information provides from virtual networks, credit card databases, e- wallet transactions, bank statements, deep web search and financial databases. Applicants profile data, unemployment histories, and income disruption histories were the elements used in the process to predict applicant's default probability risk. At this step the said social media presences (6a) or history or the person may hold views differently on the digital platforms and differently on personal levels is focused highly for the credit scoring. The information of the lifestyle of a user, expenditure of day to day life such as movies and restaurant's visits, travel details, friends list to check if any of them is already a defaulter, the abusive language used as reactions of events, sentiment analysis is run at this step. In addition to this deep web search, the unbiased details about the user are located. It provides both text and images related to the person and can also run again sentiment analysis on the same. Moreover, the said images are used to cross check with the applicant's photo where face recognition is to be run.

Then, user's financial stress data (6b) is correlated with social media presence (6a) and mathematical relationships are established between economic conditions and the applicant's financial stress. The following elements are incorporated into the financial risk score: The applicant's unemployment risk, income loss risk, income reduction risk, income variability and volatility risk. The applicant's probability of a becoming unemployed and experiencing a reduction in their income is then weighted and fetched for the applicant's financial data and incorporated into the final risk score. Said combination creates the statistical model predicting the applicant's financial stress for present and forecasted economic conditions. Said historical and forecasted economic data integrated with the applicant's profile data results in the credit scoring based ability to pay risk model and default probability utilized by the invention. Said criteria are used by lenders in making consumer credit based decisions.

Further at the same input step (6) all the affected data collected from the applicant is effectively sent in for auto filing via smart deep neural network model (14a). The applicant's application is further sent for auto filing and reviewal without human intervention via a smart deep neural network model (14a) which is based upon heuristic formulation or calculations. The data is further distributed or stored in the internal storage (14b) of a central computing system (11). This huge data gets stored in the distributed databases across the world. This is based on highly combinatorial formulation and calculation-based data mining. It includes storing and scraping of metadata as well. Also, said the stored data is secured via virtual network, financial databases, transactions, statements and a secured private key.

At the data processing step 6, data is being processed with all relevant financial information as well as other information which is obtained via an advanced processing system. All raw data are now processed into the predefined configuration. Once a person recognized his or her face than with the help of all relevant databases by financial databases data transmission are stored in the internal storage (14b). The processed data is transferred to an internal storage (14b) via a communication network module (14). Later on, this data guides for a person to seek any loan be it for the purpose of personal aspects or for the professional matters. This is all gathered along government ministry's databases which has information or details of the specific person. Thus, there will be an information filtering system which helps in making the decision for a person seeking a loan. This storage database (14b) having historical data and also stores real time data for future perspective. At this step, if the data has not been processed then the system of present invention has to follow the whole process from initial input step 1 to third input step 5. If the system of the present invention successfully completes the step 6 of data processing, then the loan automation system can proceed towards further steps of the present invention.

At input step (7) the said model takes out the various parameters as a set of inputs and executes some distinctively computative process. Loan automation system of the present invention is using computation accelerators (13) to take down heavy mathematical computation, which makes this whole process faster and smoother. Due to parallel computation run by the computation accelerators (13), also makes the faster rate of data updation.

Furthermore, at step 8, data has been transmitted from the internal storage (14b) which is generated from the past data as well as the recent approach as described in above-mentioned steps (5) and (6), where all information of a person is gathered. When the camera via video source (10) recognizes the face where his or her basic information is collected and updated automatically by smart retrieval system of the present invention. Hence, on the basis of this data, it results with the best feasible credit risk and default probability approach for decision. Further, useful data or information is transmitted from the internal storage (14b) when needed. After completing this process, an automation process is carried forward which helps in finding out best likely fit scheme.

At step 9, a credit score(9a) is generated with default probability prediction (9b) via said smart retrieval system for rating and identifying, and this score is helpful for a bank, an organization, an institute and an applicant also. As, on the filtration and suggestion based out by this process, the banks and/or the organization and/or institutes approach to the person or applicant. It can be vice versa as well, as the person or applicant itself can approach the bank, organization, and institute. Further, on the basis of this process, more financial needs or requirements are given by said search and gives discounts and offers. Hence, the said credit score (9a) and default probability prediction (9b) helps out in any loan or insurance process. Here at step 9, if credit risks score and default probability is not generated, then system of the present invention has to follow the previous step 8, in which useful data has been transmitted from internal storage (14b). If the system of the present invention successfully generates the default probability prediction and credit score, then the loan automation system can proceed towards further steps of the present invention.

The system and process of present invention are advantageous over the existing scoring and probabilities prediction process. The present invention also provides improved marketing capabilities for businesses. With the advanced prospect credit scoring via deep neural network model, marketing strategies are more focused and target the ideal populations. In addition, the present invention particularly enhances any prescreening of an individual and used to strengthen default predictions. This also allocates for the early identification of high risk individuals, narrowing the negligence probabilities and making marketing more refined. The clearly distinctive feature of present invention is most judicious combination of quantitative (financial) and qualitative parameters having end result, which is output of assessment of many more factors done by system & process reducing as well as supporting human assessment or conclusions. The present invention also preferably provides lenders with the benefits of a more accurate picture of applicant's credit risk, in both good and bad economic times; a proactive and leading indicator of credit risk and improved segmenting and delineation due to better credit risk prediction capabilities. Hence, the proposed processing system is faster, smoother than current credit risk and probability model processing systems.

While various data, records, data elements, variables and field structures of the present invention have been described in detail for the purposes of this invention, it is apparent that modification and adaptation of those elements will occur to those skilled in the art. It is expressly understood, however, that such modifications and adaptations are within the spirit and scope of the present invention as set forth in the following claims.

Claims

A process of a default probability prediction and credit scoring framework, comprising the steps of: a) fetching the user data by fetching means via video source (10); b) securing and calculating the user data via central computing unit (11) and accesses the learnt parameters via common shared memory (12); c) interfacing the said system with plurality of computation accelerators (13) for parallel computation and for faster rate of data updation; d) sending the loan application for auto filing and reviewal by deep neural network; e) creating an internal storage database (14b) by processing the user data by advanced processing system, after customer or applicant applies for a loan; f) generating inputs to smart model by intelligent retrieval system; g) making a profile after reviewing additional or outsource demographics of a person seeking loan via smart model; h) providing best feasible probability prediction of a person seeking loan or vice versa, on the basis of said credit score; i) approaching of bank, organization and institute and signing a smart or digital contract for loan disbursement; and j) monitoring and tracking the loan timely after loan disbursement.

The process of a default probability prediction and credit scoring framework as claimed in claim 1, wherein said user data includes real-time data by means of face recognition (4a), an emotion detection (4b), pulse detection (4c), speech analysis (4d) and galvanic skin response (4e).

The process of a default probability prediction and credit scoring framework as claimed in claim 1, wherein said outsource or additional demographics database of a person seeking loan includes social media presence (6a) and financial data (6b). The process of a default probability prediction and credit scoring framework as claimed in claim 1, wherein said internal storage database is being generated from the past data as well as deep network model approach.

The process of a default probability prediction and credit scoring framework as claimed in claim 1, wherein said deep neural network model is based upon heuristic formulation or calculations which takes out the various parameters as a set of inputs and executes some distinctively computative process.

6. The process of a default probability prediction and credit scoring framework as claimed in claim 1, wherein said synergic score is generated for rating and identifying the applicant's profile.

7. The process of a default probability prediction and credit scoring framework as claimed in claim 1, wherein said monitoring, and tracking of a person seeking loan is done by various sources and input to the smart neural network model.

8. A default probability prediction and credit scoring framework system, comprising:

a video source (10) in raw format for fetching user data;

a host or central computing unit (11), to store the user data, wherein said host or central computing unit (11) has computation accelerator (13), shared memory (12), and application software ;

a deep neural network model (14a), to transfer processed data to internal storage database (14b) which stores and keeps the updated data for future perspective.

9. A default probability prediction and credit scoring framework system configured to operate the steps as claimed in claims 1 - 8.