Turning prediction tools into decision tools

Arguably, the main stumbling block in getting machine learning algorithms used in practice is the fact that people do not trust them. There could be many reasons for this, for instance, perhaps the models are not sparse or transparent, or perhaps the models are not able to be customized to the user’s specifications as to what a decision tool should look like. I will discuss some recent work from the Prediction Analysis Lab on how to build machine learning models that have helpful decision-making properties. I will show how these models are applied to problems in healthcare and criminology.

Duke Scholars

Author Cynthia D. Rudin Computer Science