CMStatistics 2022: Start Registration
View Submission - CMStatistics
Title: Inference for model-agnostic longitudinal variable importance Authors:  Brian Williamson - Kaiser Permanente Washington Health Research Institute (United States) [presenting]
Susan Shortreed - (United States)
Peter Gilbert - University of Washington and Fred Hutchinson Cancer Research Center (United States)
Noah Simon - UW Biostatistics (United States)
Marco Carone - University of Washington (United States)
Abstract: In many applications, it is of interest to assess the relative contribution of features (or subsets of features) toward the goal of predicting a response; in other words, to gauge the variable importance of features. We will discuss a general framework for nonparametric inference on interpretable algorithm-agnostic variable importance, where variable importance is defined as a population-level contrast in oracle prediction potential between two nested groups of features. In this framework, valid confidence intervals and tests may be constructed, even when machine learning techniques are used. We will further discuss several approaches to summarizing the longitudinal importance of variables and to making inferences from these summaries.