A Simulation Study on Group Variable Selection Methods for Binary Response

Abstract

     Binary data, denoting data having two alternative outcomes, is frequently observed across several research domains including finance, social sciences, psychology, and health. The logistic regression model is extensively employed for the analysis of binary data. It is essential to meticulously examine the detection and management of influential outliers to guarantee the suitability of the fitted binary logistic models. This article offers an extensive evaluation of various collective binary logistic techniques employed in regression models, emphasizing a comparison of the efficacy of four distinct logistic regression approaches. The methods encompass group Lasso binary estimates, group mcp binary estimates, group scad binary estimates, and binary regularization paths for generalized linear models by coordinate descent (glmnet) estimates. The comparisons derive from a simulation research aimed at determining which of these approaches exhibits superior performance across all regression scenarios. This review and comparison enable researchers and practitioners to discern the most effective methodologies for evaluating binary data via logistic regression.