The Group Lasso-Ridge Hybrid Method to Selection of Variables for Smoker in Jambi Province
Rini Warti (a*), Khairil Anwar Notodiputro (b), Bagus Sartono (b)

a) Mathematics Education Department, UIN Sulthan Thaha Saifuddin, Jambi, 36363, Indonesia
* riniwarti[at]uinjambi.ac.id
b) Statistics Department, IPB University, Bogor, 16680, Indonesia

Abstract

The 2018 RISKESDAS report by the Health Research and Development Agency at the Indonesian Ministry of Health states that the number of smokers in Indonesia tends to increase yearly. Even the prevalence of smokers over the age of 15 is relatively high, with the number of smokers at 33.8%. Many factors influence people to smoke. The factors observed consisted of 14 groups and 30 variables. One method used to select variables in group form is the Group Lasso. This study aimed to determine a predictor variable from a binary response variable using the Group Lasso-ridge hybrid method for smokers in Jambi Province. The data was taken from the 2017 Indonesian Demographic and Health Survey (IDHS), classifying smokers into two categories (smoking or not). Data analysis consists of two stages. The results showed that of the 30 predictor variables used in the first stage, 18 variables were selected that affected people^s smoking, with a predictive performance score of 68.09. All variables selected in step one will use in the second stage. In stage two, there are five variables were set that influenced people to smoke from the 18 variables used, with a predicted performance score of 74.99

Keywords: Group Lasso, Group Lasso-ridge hybrid, Smoker, Variable selection

Topic: Mathematics