Time: 16:00-17:00, Thursday, July 11 2024
Venue: E4-233, Yungu Campus
Host: Zhennan Zhou, ITS
Speaker: Guoyi Zhang, University of New Mexico
Title: Small Area Estimation using Support Vector Machine
Abstract: The Fay-Herriot model (Fay&Herriot,1979) and the Battese, Harter, and Fuller (BHF) model (Battese, Carter, &Fuller,1988) are commonly used in small area estimation (SAE), assuming a linear relationship between response and predictor variables, which may not always hold. This study extends both models to more flexible semi-parametric versions (SP-SAEArea and SP-SAEUnit) where nonparametric components are estimated using machine learning tools like support vector machines. We propose two backfitting algorithms to address the SP-SAEArea and SP-SAEUnit problems. Using American Community Survey (ACS) data and simulations, our SP-SAEArea and SP-SAEUnit models significantly enhance estimation accuracy by simultaneously reducing bias and variance when linear relationships are not met. This research underscores the importance of integrating machine learning techniques into small area estimation to improve accuracy by addressing model assumptions.
