A risk rating is a numerical estimate of the risk associated with a borrower or other subject. In credit decisions, an ML system produces the rating before approving or rejecting an application. It may be binary, such as 1 or 0 at a chosen threshold, or continuous between 0 and 1. section 2.4
Credit approval
A bank weighs two asymmetric mistakes: lending to a likely defaulter and denying a loan to someone who would repay. Comparing people, credits, or money lent—and checking whether the tradeoff changes over time—helps set useful requirements and metrics. section 2.4
Use a risk rating to frame decisions around:
- The errors the system must avoid. section 2.4
- The business cost of each type of mistake. section 2.4
- The metrics and data needed to support a successful system. section 2.4
