Modification of the MDR-EFE method for stratified samples

TL;DR

This paper adapts the MDR-EFE method for stratified samples with binary outcomes, establishing consistency criteria and demonstrating advantages over traditional i.i.d. approaches through analysis and simulations.

Contribution

The paper introduces modifications to the MDR-EFE method tailored for stratified samples, including a new consistency criterion and a cost approach for experiment comparison.

Findings

Proves strong consistency of the modified method

Demonstrates improved performance in stratified samples

Validates advantages through simulations

Abstract

The MDR-EFE method of performing identification of relevant factors within a given collection X_1,...,X_n is developed for stratified samples in the case of binary response variable Y. We establish a criterion of strong consistency of estimates (involving K-cross-validation procedure and penalty) for a specified prediction error function. The cost approach is proposed to compare experiments with random and nonrandom number of observations. Analytic results and simulations demonstrate advantages of the method introduced for stratified samples over that employed for i.i.d. learning sample.