Title: A hierarchical model to handle missing data in high-dimensional spatio-temporal models
Authors: Shyam Ranganathan - Virginia Tech (United States) [presenting]
Abstract: High-dimensional spatio-temporal models have important applications in health analytics. A number of variables, especially from birth records are available with increasing frequency, and a current challenge is to make efficient models that can explain the data in a manner that corresponds to causal or mechanistic explanations (as opposed to black box machine learning methods). We demonstrate how missing data can significantly impact inferences and prediction, and hence black box methods are ``dangerous''. We present an efficient model to handle missing data in these complex settings and present a hierarchical model that is better suited to prediction in these high-dimensional spatio-temporal problems.