2018-2019 exchange Winter 2019 Newsletter FINAL | Page 14

Glenn: To me, data science and data analytics are the same. When
people think of data analytics, they understand it in its simplest form of
just creating a report or dashboard that business leaders can look at it
and draw some conclusions. We call that descriptive analytics, or
business intelligence.
With predictive analytics, you’re not just describing the data and creat-
ing some nice charts, but you’re predicting what will happen next.
What will people do and how will people react to whatever is there.
Things like insurance risk— what’s the likelihood of people dying or
having an accident?
And then there is the prescriptive analytics side. After we have
predicted what might happen, what’s next is what are we going
to do about it. Is there something we can do to change people’s
behaviors? Then there is pre-emptive analytics, which is how
can we prevent something from happening that might otherwise
happen, like fraud.
People might hear about things like neural networks or deep learning
or regression or time series models, which are called supervised
modeling methodologies. A supervised model means that you have a
target that you are trying to predict and you have past data on that
target. Something like marketing response or a fraud you knew was
there, or a credit default. You know what you are predicting and you
have some data to predict, which is usually historical data and you
build a predictive model that addresses that target.
Jen: One of the books I’ve read was about “separating the signal
from the noise”, which was about predictions and the mentality of
“more is more” when it comes to data. At some point, having too
much data can be an obstacle to identify the insights you are
looking for.
Glenn: That’s the essence of the science of statistics. On a high
level, it all depends on the signal one is trying to detect. Take a
classic example: if you have a coin, is it a fair coin? Does it land
heads and tails 50% of the time usually? Or is it way off- 70-30 be-
cause it’s oddly shaped. Say you want to distinguish the 50-50 from
the 70-30. If you cast that coin 10-15 times, it’s probably good
enough to figure that out. But if you were to detect that the coin is 51
14