System, method and computer program for underwriting and processing of loans using machine learning
A system and method for processing loans includes loan approval decision module that receives input from a loan applicant and collects external data including credit bureau data, bank transaction data, and social media data. The system also includes a machine learning module having a pre-processing subsystem, an automated feature engineering subsystem and a feature statistical assessment subsystem. A business objective determination module and an adverse notice notification module is also provided. The business objective determination module includes a weight optimization company valuation maximization model. A set of models is developed using the machine learning module to predict performance of the borrower based on the business objective determination.
1. A method comprising:
receiving data indicative of a loan application from a loan applicant;
collecting, based on receiving the data indicative of the loan application, external data related to the loan applicant;
pre-processing the external data to generate processed external data;
conducting an automated feature engineering to develop a set of features;
determining, based on the processed external data, an objective function by calculating values for a plurality of variables representing corresponding business objectives and assigning weights to the plurality of variables;
creating an ensemble machine learning model from a plurality of machine learning models, wherein at least a portion of the plurality of machine learning models are one or more of trained or selected based on optimizing the objective function;
varying one or more of the weights over time to account for changes in one or more business conditions;
updating the ensemble machine learning model based on changes in the objective function corresponding to the varying of the one or more weights;
determining, based on applying a localized linearity process to the ensemble machine learning model, one or more categories of reasons for rejection mapped to adverse action notices; and
sending, to a computing device associated with the loan application and based on the mapping of the one or more categories to the adverse action notices, an indication of an adverse action notice.
2. The method of claim 1 , wherein the automated feature engineering transforms the external data through scaling, decomposition and aggregation.
3. The method of claim 1 , wherein the pre-processing comprises formatting, cleaning, and sampling the external data.
4. The method of claim 1 , wherein the optimizing the objective function is based on maximizing the value of a lender associated with the loan application.
5. The method of claim 1 , wherein the plurality of variables representing corresponding business objectives comprises one or more of:
a first payment default recovered,
a return on capital,
a cost of customer acquisition,
a cost of maintaining a customer, or
a customer lifetime value.
6. The method of claim 1 , wherein conducting automated feature engineering comprises generating new features, and determining which algorithms require feature engineering.
7. The method of claim 1 , wherein applying the localized linearity process to the ensemble machine learning model comprises linearly approximating, for a particular loan applicant, a non-linearity associated with the ensemble machine learning model to identify the one or more categories of reasons for rejection.
8. The method of claim 1 , wherein the indication of the adverse action notice is in compliance with one or more regulations without requiring human review of the adverse action notice.
9. A non-transitory computer-readable medium storing computer-executable instructions that, when executed by one or more processors, cause:
receiving data indicative of a loan application from a loan applicant;
collecting, based on receiving the data indicative of the loan application, external data related to the loan applicant;
pre-processing the external data to generate processed external data;
conducting an automated feature engineering to develop a set of features;
determining, based on the processed external data, an objective function by calculating values for a plurality of variables representing corresponding business objectives and assigning weights to the plurality of variables;
creating an ensemble machine learning model from a plurality of machine learning models, wherein at least a portion of the plurality of machine learning models are one or more of trained or selected based on optimizing the objective function;
varying one or more of the weights over time to account for changes in one or more business conditions;
updating the ensemble machine learning model based on changes in the objective function corresponding to the varying of the one or more weights;
determining, based on applying a localized linearity process to the ensemble machine learning model, one or more categories of reasons for rejection mapped to adverse action notices; and
sending, to a computing device associated with the loan application and based on the mapping of the one or more categories to the adverse action notices, an indication of an adverse action notice.
10. The non-transitory computer-readable medium of claim 9 , wherein the automated feature engineering transforms the external data though scaling, decomposition and aggregation.
11. The non-transitory computer-readable medium of claim 9 , wherein the pre-processing comprises formatting, cleaning, and sampling the external data.
12. The non-transitory computer-readable medium of claim 9 , wherein optimizing the objective function is based on maximizing the value of a lender associated with the loan application.
13. The non-transitory computer-readable medium of claim 9 , wherein
the plurality of variables representing corresponding business objectives comprises one or more of:
a first payment default recovered,
a return on capital,
a cost of customer acquisition,
a cost of maintaining a customer, or
a customer lifetime value.