Predictors of COVID-19 vaccination rate in USA: A machine learning approach

Syed Muhammad Ishraque Osman; Ahmed Sabit

doi:10.1016/j.mlwa.2022.100408

Predictors of COVID-19 vaccination rate in USA: A machine learning approach

Mach Learn Appl. 2022 Dec 15:10:100408. doi: 10.1016/j.mlwa.2022.100408. Epub 2022 Sep 16.

Authors

Syed Muhammad Ishraque Osman¹, Ahmed Sabit²

Affiliations

¹ Jack Welch College of Business & Technology, Sacred Heart University, West Campus, East Building - 1st Floor, 3135 Easton Turnpike, Fairfield, CT 06825, United States of America.
² Department of Biostatistics, The Johns Hopkins University, 615 North Wolfe Street, Baltimore, MD 21244, United States of America.

Abstract

In this study, we examine state-level features and policies that are most important in achieving a threshold level vaccination rate to curve the effects of the COVID-19 pandemic. We employ CHAID, a decision tree algorithm, on three different model specifications to answer this question based on a dataset that includes all the states in the United States. Workplace travel emerges as the most important predictor; however, the governors' political affiliation (PA) replaces it in a more conservative feature set that includes economic features and the growth rate of COVID-19 cases. We also employ several alternative algorithms as a robustness check. Results from these checks confirm our original findings regarding workplace travels and political affiliation. The accuracy under different model specifications ranges from 80%-88%, whereas the sensitivity is between 92.5%-100%. Our findings provide actionable policy insights to increase vaccination rates and combat the COVID-19 pandemic.

Keywords: COVID-19; Decision tree; Health policy; Machine learning; Vaccination; Vaccine hesitancy.