search for




 

Comparison of penalized zero inflated negative binomial regression methods
Journal of the Korean Data & Information Science Society 2021;32:715-37
Published online July 31, 2021;  https://doi.org/10.7465/jkdi.2021.32.4.715
© 2021 Korean Data and Information Science Society.

Hye-Yeon Sin1 · Joonsung Kang2

1Department of Pharmacy, Duksung Women’s University
2Department of Information Statistics, Gangneung-Wonju National University
Correspondence to: This study was supported by 2018 Academic Research Support Program in Gangneung-Wonju National University.
1 Associate professor, Department of Pharmacy, Duksung Women’s University, Samyang-ro 144-gil 33, Dobong-gu, Seoul 01369, Korea.
2 Associate professor, Department of Information Statistics, Gangneung-Wonju National University, Jukheon-gil 7, Jibyeon-dong, Gangneung-si, Gangwondo 25457, Korea.
E-mail: mkang@gwnu.ac.kr
Received May 3, 2021; Revised May 31, 2021; Accepted June 11, 2021.
This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/3.0) which permits unrestricted non-commercial use, distribution, and reproduction in any medium, provided the original work is properly cited.
Abstract
We take Poisson regression and negative binomial regression dealing with a count data as a response variable into account. The negative binomial regression considers overdispersion more than the Poisson regression. On the other hand, with a high frequency of 0 values, zero inflated count model is preferred. In the paper, as we consider variable selection, frequent 0 values, and overdispersion, a penalized zero inflated negative binomial regression is applied. In numerical studies, we compare penalized zero inflated negative binomial regression methods for count data by using those penalties. The simulation studies are conducted to show the finite sample performance of various methods by providing root mean square error and prediction accuracy measure. Real data analysis is also demonstrated for illustration purpose. Numerical studies show that each penalized method has almost similar root mean square error and prediction accuracy measure in the zero inflated overdispersed data. Mnet method has relatively lower root mean square error and higher prediction accuracy.
Keywords : Adaptive Lasso, mnet, Poisson regression, snet, zero inflated negative binomial regression.