CFE 2019: Start Registration
View Submission - CMStatistics
Title: Variable selection and prediction in logistic regression with incomplete data Authors:  Rong Xing - Shanghai University of International Business and Economics (China) [presenting]
YunXiang Cao - Shanghai University of International Business and Economics (China)
Abstract: Multiple imputation random lasso (MIRL) method is an extension of the random lasso to combine penalized regression techniques with multiple imputation techniques. This paper aims to apply MIRL to logistic regressions in order to deal with classification problem with incomplete high-dimensional predictors. The missing completely at random pattern is considered. Extensive simulation studies are conducted to compare MIRL-logistic with its several alternatives. The result shows that MIRL-logistic has an improved accuracy on both outcome prediction and variable selection performance in high-dimensional scenarios. Especially, the proposed method performs well when the correlation among the predictors is high.