Detecting fraud in credit applications
The concept involves efficiently using machine learning to quickly identify possible fraudulent applications in small business loan and credit applications by automatically flagging applications that meet certain criteria. In one preferred implementation, the tool compares a business description to a selected NAICS code in a loan application to assess the potential for fraud. Specifically, an algorithm can match the leftmost two digits of the selected code with the description of the category from an applicant. An engine calculates a probability of a fraud score based on the matching attached to the application. Because the tool detects fraud proactively rather than reactively, it substantially reduces computational costs and resources and reduces the biases associated with highly intensive manual work.
1. A computer system for processing loan applications, comprising:
one or more processors; and
non-transitory computer-readable storage media encoding instructions which, when executed by the one or more processors, causes the computer system to:
create a classification model by training a neural network with a plurality of layers using a training data set;
weight classes within the classification model to facilitate learning across the classes;
receive a business description of a business applying for a loan from an applicant;
generate, using the classification model, classification options for the business using the business description;
compare a selection of one of the classification options by the applicant to the classification options;
determine if a fraud threshold is met based upon the compare;
when the fraud threshold is met, identify the loan as problematic; and
when the fraud threshold is not met, identify the loan as nonproblematic.
2. The computer system of claim 1 , wherein the loan is a Payroll Protection Program loan.
3. The computer system of claim 2 , wherein the classification options are based upon North American Industry Classification System codes.
4. The computer system of claim 2 , comprising further instructions which, when executed by the one or more processors, causes the computer system to use a leftmost two digits of the selection to determine the fraud threshold.
5. The computer system of claim 2 , wherein the business description is less than 100 characters in length.
6. The computer system of claim 1 , wherein the loan application is for a credit card product.
7. The computer system of claim 1 , wherein the fraud threshold is adjustable.
8. The computer system of claim 1 , wherein the classification options are based upon North American Industry Classification System codes.
9. The computer system of claim 1 , comprising further instructions which, when executed by the one or more processors, causes the computer system to use a leftmost two digits of the selection to determine the fraud threshold.
10. The computer system of claim 1 , wherein the business description is less than 100 characters in length.
11. A computer-implemented method capable of processing loan applications comprising:
creating a classification model by training a neural network with a plurality of layers using a training data set;
weighting classes within the classification model to facilitate learning across the classes;
receiving a description of a business applying for a loan from an applicant;
generating, using the classification model, classification options for the business using the description;
comparing a selection of one of the classification options by the applicant to generated options;
determining if a fraud threshold is met based upon the comparing;
when the fraud threshold is met, flagging the loan as problematic; and
when the fraud threshold is not met, identifying the loan as nonproblematic.
12. The method of claim 11 , wherein the database is sourced from third-party sources.
13. The method of claim 11 , wherein a certain percentage of applications are flagged based on the comparing.
14. The method of claim 13 , further comprising modifying the comparing to accommodate detection of a high-risk industry.
15. The method of claim 11 , further comprising applying the comparing to government loan programs.
16. The method of claim 11 , further comprising allowing the applicant to select a sub-industry option.
17. The method of claim 11 , further comprising allowing the applicant to apply online from a client device.
18. The method of claim 11 , wherein the fraud threshold is set to a confidence level.
19. A computer system capable of detecting fraud in Payroll Protection Program loans submitted to a financial institution, comprising:
one or more processors; and
non-transitory computer-readable storage media encoding instructions which, when executed by the one or more processors, causes the computer system to:
create a classification model by training a neural network with a plurality of layers using a training data set;
weight classes within the classification model to facilitate learning across the classes;
receive text input from a customer associated with an application for a Payroll Protection Program loan, wherein the text input is a description of a business;
generate, using the classification model, classification options for the text input using the description of the business, wherein the classification options are presented to the customer on a graphical user interface;
compare a selection of one of the classification options by the customer to generated options, wherein the customer manually elects a classification options which is not one of the generated options; and
based on a rating score, flag the application for review by the financial institution.
20. The system of claim 19 , wherein the rating score is binary.