Page 8 - spark_ezine

SPARK 2019 CONFERENCE

ANALYTICS
Improving

Performance.

Building better solutions with explainable AI

BY: CHRIS YASKO, DATA SCIENCE LAB, RAJ BONDUGULA, PH.D, DATA SCIENCE LAB,
AND MATT TURNER, PH.D, DATA SCIENCE LAB

EDITOR’S NOTES

It’s no secret in the data science community that machine learning techniques can yield superior
results as compared to traditional methods. The challenge has been that data enters one end,

results come out the other but it’s not clear what’s driving the results. In short, it’s a black box
and not usable in the highly regulated credit-risk industry. That is until recently, when Equifax
Senior Data Scientists came together in a singular effort to crack the black box of ML/ AI
algorithms.

Our first patent was NeuroDecision Technology and since then we’ve applied for a series of
®
new patents around explainable AI. For the first time, we now have the capabilities to uniquely
create reason codes for models that are compliant, logical, and actionable.

A GROWING PIPELINE OF INVENTIONS

NeuroDecision Synthetic Online Binary Trees Gradient Configurable Text Classification &
Generation 2 Entity Detection Boosted Machines Solutions Seed Data

Explainable machine Detect synthetic identities in Explainable machine Create and manage Improve efficiency of
learning method using online transactions learning method for tree automated modeling text classification using
neural networks with a based models systems, particularly automatically generated
single hidden layer segment-specific seed data
executable programs

Nov 2017 Nov 2017 Nov 2017 Nov 2017 May 2018
WWW.EQUIFAX.COM/SPARK