An application of a multiple neural network learning system to emulation of mortgage underwriting judgements

Abstract

A multiple neural network learning system (MNNLS) was used to replicate the decisions made by mortgage insurance-underwriters. The MNNLS was trained on previous underwriter judgements and learned to mimic their underwriting skills. The system reached a high degree of agreement with human underwriters when testing on previously unseen examples. Disagreements were examined using case studies, a single feature distribution analysis and a quality analysis. These studies indicate that human underwriters in many cases disagree with one another and are inconsistent in the use of their underwriting guidelines. It was found that when the MNNLS system and the underwriter disagree, the system's classifications are more consistent with the guidelines than the underwriter's judgement.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>