Predictive analytics: Validating the black box

“Reveal hidden patterns in your data” and “Transform your data into meaningful insights” are phrases we often hear when it comes to predictive modelling. However, these hard sell phrases tend to lack substance and what remains is a black box. We discuss various techniques where we evaluate the decisions made by these automated algorithms and how we can adopt a more transparent modelling approach. We demonstrate how evaluation techniques not only lift the lid on the ‘black box’ but also allow us to more accurately assess the risk involved. This enables us to deliver a fair and accurate price to our customers.