Regency EF
Subsemble Estimation for Spatial Count Data (304035)
*Aimee Schwab-McCoy, Creighton UniversityKeywords: Subsemble estimation, count data, spatial models
Subsemble estimation is a subset-based ensemble prediction method which partitions a data set into subsets of observations, fits a model or prediction algorithm on each subset, and uses k-fold cross-validation to output a prediction. Subsemble estimation has been shown to have good prediction performance and some attractive theoretical qualities (Sapp, van der Laan, & Canny, 2014), and improves computational speed in spatial models (Barbian & Assuncao, 2017).
Previous applications and simulation studies of subsemble estimation have focused on the Gaussian data case. This poster will present results of a simulation study testing the performance of subsemble estimation on spatially distributed count data – specifically Poisson and beta-binomial response models. We will provide recommendations for effective weighting schemes and re-combination algorithms to use in the k-fold cross validation step, as well as suggestions on improving overall performance in both parameter estimation and prediction accuracy.