You are on page 1of 2

The Next Big Step in Machine Learning

By: Stephanie Krutick, Commercial Analytics Leader

Equifax Inc.
As originally published in the Credit Research Foundation 2Q 2018 CRF News

Think neural networks in today’s fast-moving networks are rather complex and the outcomes
environment - machine learning and data are not easily explainable to consumers, which
analytics innovations are helping organizations has raised compliance concerns.
efficiently and intelligently use big data to Boosting interest in the data analytics
improve business outcomes. Neural networks, industry
often used within more progressive forms of In the credit risk industry, there is an
machine learning known as “deep learning,” appetite to explore new machine learning
can push past the linear boundaries of logistic approaches. Imagine being able to
regression analysis, opening up a new world of improve approval rates by 7%* by using
opportunity with their more precise predictive machine learning techniques like neural
potential. Let’s look at why this is great news for networks. Excitement around the potential of
businesses and the consumers they serve. neural networks and related machine learning
technologies is mounting. Experts within the
Explaining straight line versus an inclusive arc data science and analytics worlds have been
When used for credit assessment, traditional pressing hard for a solution that’s applicable for
logistic regression models [think: everyone use in credit lending decisions. As a result, many
above the straight line is approved and everyone firms are producing powerful tools using deep
below is declined] tend to put consumers into learning techniques.
two categories, such as prime and sub-prime.
This can lead to the exclusion of individuals Leading firms in data & analytics have developed
who are suspended just below prime as they a game-changing technology that leverages
work toward building (or improving) their credit the advanced, machine-learning technology of
profile. One reason these models are popular neural networks for risk decisioning. This new
is because the results can be easily explained technology is highly configurable, granular and
to consumers: supporting compliance with predictive. Like traditional models, it offers
Consumer Financial Protection Bureau (CFPB) actionable reason codes that easily explain the
consumer credit regulations that require resulting scores, which supports regulatory
increased lending transparency. For example, requirements.
in a logistic regression framework, a factor
like the level of income may only be a positive The unique, patent-pending technology
contributor to a score, while in a neural network facilitates a deeper learning of a consumer’s
scenario it could sometimes be a positive behavior and can be applied and deployed
contributor to a score, and in other scenarios wherever traditional scorecards are appropriate
be a negative contributor to a score, without - across both the commercial and consumer
an explanation as to why. This may lead to segments.
compliance or regulatory risk.
The results speak for themselves
Neural networks create a curved, non- Market results show a 7% increase in approval
linear arc that can help increase the pool of rates* and a 13% increase in detecting bad
“approved” consumers by capturing hard-to- customers*over and above logistic regression
score consumers who are not quite prime, yet techniques, improving profitability. The benefits
are clearly trending in the right direction. This can be realized in the knowledge that the scores
could benefit businesses by helping them more are explainable and that reason codes can be
securely serve these consumers and drive provided to consumers or businesses when
growth. Neural networks can benefit consumers requested.
by giving more people access to mainstream
financial services. The problem is, these

©2018 Credit Research Foundation

More to come in the near future About the Author:
For data scientists within the credit and lending Stephanie Krutick is the
industries, this technology is game-changing. Commercial Analytics
Businesses can finally create a high-performing Leader at Equifax. Her team
model that maximizes predictiveness and is responsible for driving
accuracy, while also meeting critical regulatory new product innovation
guidelines. However, if recent advances in and custom analytics in
neural networking and machine learning tell the commercial risk and
us anything, it’s that we can expect even more marketing space. Stephanie
advances in the near future. This is just the holds a B.S. in Mathematics
beginning of many exciting developments ahead. and a Master’s in Applied
Statistics from Kennesaw State University.
* Based on a Commercial Insight Delinquency
Score improvement achieved by Equifax in 2017.

©2018 Credit Research Foundation