How to make predictive models good (and accurate)
Kareem Carr, guest-posting at Mathbabe.org
All three principles have one underlying idea. Bad data science obscures and ignores the real world performance of its algorithms. It relies on little to no validation. When it does perform validation, it relies on canned approaches to validation. It doesn’t critically examine instances of bad performance with an eye towards trying to understand how and why these failures occur. It doesn’t make the nature of these failures widely known so consumers of these algorithms can deploy them with discernment and sophistication.
Thomas Lumley (@tslumley) is Professor of Biostatistics at the University of Auckland. His research interests include semiparametric models, survey sampling, statistical computing, foundations of statistics, and whatever methodological problems his medical collaborators come up with. He also blogs at Biased and Inefficient See all posts by Thomas Lumley »