Robust inference with incompleteness for logistic regression model

Abstract

Logistic regression models traditionally assume observed covariates. However, practical scenarios often involve missing data and outliers, which pose significant challenges. This short communication presents a new approach to solve these issues by integrating random covariates following a Student $t$-distribution within the framework of logistic regression. We propose a Robust Stochastic Approximation Expectation–Maximization algorithm suitable for Logistic Regression (REM-LR) that, in addition, is able to improve the resilience of the model against both missing values and outliers.