Imposing Domain Expertise on Algorithmic Learning to Construct Highly Predictive and Palatable Scorecards
2013 INFORMS Annual Meeting Minneapolis
Innovative Applications in Analytics Award Semi-Finalists
Presented by Gerald Fahner, FICO
Credit scoring systems must be easy to explain and conform to certain legal/operational constraints. Algorithmic learners such as Tree Ensemble Models (TEMs) often outperform traditional scorecards on predictive grounds, but may not satisfy all the constraints. In this session, Fahner describes the novel approach to productively exploit valuable information found by TEMs to inform the construction of compliant scorecard systems with superior predictive power as compared to traditional scorecard development approaches.
