A0211
Title: Switching meta-regression model in high dimensional data
Authors: Ivy Corazon Ancog - Bohol Island State University and University of the Philippines (Philippines) [presenting]
Erniel Barrios - University of the Philippines (Philippines)
Joseph Ryan Lansangan - University of the Philippines (Philippines)
Abstract: The aim is to estimate a semiparametric mixed switching meta-regression model with high dimensional predictors in a hybrid of cubic smoothing splines (fixed effect component) and a Restricted Maximum Likelihood (REML) (random effect component) embedded in the backfitting algorithm. Sparse Principal Component Analysis is used in dimension reduction and variable selection but a nonparametric function of the sparse principal components is postulated in the model to address the decline in predictive ability of the model due to the use of only few components of the high dimensional predictors. We also consider regime switch (two possible groupings) identified via Support Vector Machine Classifier. Simulation study shows that the proposed procedure yields better predictive ability (Mean Absolute Percentage Error) than the Ordinary Least Square (OLS) method.
