The election of Donald Trump, the Brexit vote, the impeachment of the Brazilian president. In a different time,...
By submitting your personal information, you agree that TechTarget and its partners may contact you regarding relevant content, products and special offers.
all these things would have looked fairly improbable, but they were par for the course in an unpredictable 2016.
"It points out that we have to be constantly looking for changes in the world that might affect our models," said Tom Davenport, founder of the International Institute of Analytics, based in Portland, Ore., in a webcast. "As we become more reliant on machine learning, coming to terms with changes in the world that affect our models will be very important."
He said all the upheaval of the previous year has relevance for businesses that are trying to predicate aspects of their operations on predictive analytics algorithms. The unpredictable nature of events means enterprises can't get complacent.
There may be a tendency to trust a model once it starts working, but it became apparent this year that social trends can change quickly. It's important for businesses to constantly interrogate their models and adjust assumptions upon which they are built, Davenport said.
"We need to check these things occasionally to see that they're working," he said.
For Forrester analyst Mike Gualtieri, the situation points out the need for analytics professionals to be aware of cognitive biases. He said most of the pollsters and election forecasters who predicted big margins of victory for Hillary Clinton relied on relatively weak polling data. But it confirmed what everyone thought they knew to be true, so no one really questioned it.
Gualtieri said this is similar to how a lot of supposedly data-driven decisions are made. Analyses that confirm existing opinion are used, while those that challenge assumptions are disregarded. In the year ahead, businesses should learn from these forecasting missteps and work to overcome the cognitive biases in their own organizations.
"Between the Brexit and Trump getting elected, maybe that was a sign of complacency at different levels," he said.
For Pradeep Mutalik, associate research scientist at Yale, part of what's making events so unpredictable today is passion. He said whether we're talking about the Brexit vote or Trump's election, voters seem to have been driven by passions, which are hard to measure in predictive models. This is adding a new layer of unpredictability to world events, and it's something businesses need to account for when developing predictive analytics algorithms.
"It started with Brexit," he said. "It seems to happen when one side is very passionate, compared to the other side."
Predictive modeling demands mix of many different skills
Building predictive models may seem glamorous, but it should be about business value
Speeding up predictive modeling efforts can lead to bigger payoff