Machine learning for time-to-event data and its relationship to the Kaplan-Meier estimator

Professor Richard Hahn
Arizona State University

We review aspects of time-to-event data that make it challenging for non-parametric covariate adjustments. Recent work using tree-ensembles for survival analysis is based on a covariate-dependent version of the traditional Kaplan-Meier estimator, but is computationally impractical in many applied settings. We derive a continuous-analogue of the Kaplan-Meier estimator and describe how it can be used to do survival analysis with generic machine learning methods for conditional density estimation. 

Speaker Biography: P. Richard Hahn is Associate Professor of Statistics at Arizona State University. Prior to that he was Associate Professor of Econometrics and Statistics at the University of Chicago Booth School of Business. He received his PhD in Statistics from Duke University in 2011 and he also holds a bachelor's degree in Philosophy of Science from Columbia University. His research interests include Bayesian machine learning for causal inference, Monte Carlo methods, decision theory, interpretable data science, and semi-supervised learning. He escapes Phoenix's scorching summers in the cool mountains of Southwestern New Mexico with his wife and two young children.