Machine learning Bureau score for Home Lending in an American finance company

Our client is a leading provider of mortgage financing, originating loans and lines of credit to consumers in the US. Currently, they receive applications where applicants provide personal information and a soft pull of their FICO score is requested. That score is used to evaluate the applicant’s credit worthiness and determine conditional approval and the type of product available for the customer, including conventional, FHA or other mortgage loans. After conditional approval, a formal application is initiated, and underwriters review the information to determine the final application decision. When evaluating applications below regulatory and business thresholds, the company has the intention to approve more applications and increase loan volume, and there is an expectation that through the enhanced credit assessment, our client will improve the percentage of Low to Moderate Income (LMI) population able to obtain mortgage loans. Both aspects have a direct impact on the reputation and economic profits of the firm, so they are of pressing importance to the company. This project aims to build an applicant-level bureau-only score based on upgraded bureau internal attributes. This score will eventually serve as the basis for evaluating a customer’s credit risk before any loan structure or collateral information is considered. It will be used as a standalone score that can be used in the initial customer evaluation to identify better leads (mortgage inquiries for preapproval) and as input to a future application-level model.