Big Data Meets Human Understanding : Interpretability in Predictive Modeling with the Bayesian List Machine

• Activity Number: 328
• Type: Invited
• Date/Time: Tuesday, August 6, 2013 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistical Computing
• Author(s): Benjamin Letham and Cynthia Rudin and Tyler H. McCormick and David Madigan*+
• Companies: MIT and Massachusetts Institute of Technology and University of Washington, Seattle and Columbia University
• Keywords: power grid ; machine learning ; maintenance ; supervised ranking

Abstract

Interpretability is an important quality in a predictive model. In this talk, I will describe how we are are designing models consisting of a series of if...then... rules that discretize a high-dimensional, multivariate feature space into a series of simple, readily interpretable decision statements. Specifically, I will introduce the "Bayesian List Machine," and discuss its application to big data in medicine. In particular, I will provide an alternative to the CHADS2 score, which is a widely used scoring system for stroke prediction.