Explainable AI and the FICO Score

Explainable AI — or algorithmic transparency — is a hot topic in predictive analytics, even in the realm of credit scoring. This is what I’ll address in my presentation on November 19 at the International Conference on Data Analytics in Athens.

My presentation on “Transparency by Design: An Explainable AI Approach to Credit Risk Scoring” discusses our application of recent machine learning technologies to challenge the US FICO® Score. It turns out that highly predictive models don’t need to be black boxes, at least not in credit scoring.

I’ll also explain a streamlined analytic pipeline FICO has developed to generate highly predictive and transparent generalized additive model trees that synthesize data-driven learning with domain expertise to create explainable analytics. As I noted in a previous post, off-the-shelf algorithmic learners have severe limitations: The modeled relations may be unpalatable. The procedures do not allow imposing domain expertise or constraints into these models. This is a big problem in a highly regulated area like credit risk scoring; legal and operational constraints may prevent direct deployment of these models.

Our innovative approach reconciles the power of algorithmic learning with practical constraints, to effectively develop superior, palatable scorecards that can be practically deployed. The new approach consists of two stages:

1. Learn and diagnose a purely data-driven, algorithmic model of the data. This is great for gaining objective insight into the empirical relation, and yields a highly accurate fit to the historic data. However, the result may violate certain business constraints.
2. Transmute the algorithmic model, at minimal loss of information, into a (segmented) scorecard model, whereby constraints on score weights can be imposed, as desired, to adhere to legal and operational constraints.

We like this new approach because it allows us to build better models faster, it produces explainable AI, and we frequently see significant lift over traditional model developments.

For more information, see my previous post on this approach, and Ethan Dornhelm’s post, Can Machine Learning Build a Better FICO Score? If you’re traveling to the conference, introduce yourself and let’s talk!