Everyone into the (Data) Pool
Predictive models are hungry for data, but adding features to a model won’t necessarily improve its performance or justify the cost.
Read More
In early 2023, our data science director Michael Niemerg led a study to determine whether adding hundreds of consumer data features to our Curv® Group Health model would improve its performance. The results—published as a Milliman white paper—came as a surprise to many readers who are used to hearing that predictive models have a nearly insatiable appetite for data.
That research prompted Michael to write an article for The Actuary, explaining why adding features may not improve model performance. Worse, they could add cost, reduce explainability, or even inject bias.
His article, “Everyone Into the (Data) Pool,” also outlines a framework that modelers can use to help decide when new features are worthwhile, with return on investment being a key metric. It seems that in predictive modeling, as in life, “work smarter, not harder” is good advice. Read it for a peek into the high-level thinking that informs our innovative insurtech.
