Central limit theorem related to MDR-method

TL;DR

This paper extends the MDR-method for high-dimensional biological data by introducing regularized estimates and proving a multidimensional CLT, enhancing the statistical understanding of factor significance in binary response models.

Contribution

It introduces regularized versions of prediction error estimates in the MDR-method and establishes their multidimensional CLT, advancing statistical inference in high-dimensional settings.

Findings

Proved multidimensional CLT for regularized estimates

Discussed self-normalization variants of the CLT

Established strong consistency conditions for estimates

Abstract

In many medical and biological investigations, including genetics, it is typical to handle high dimensional data which can be viewed as a set of values of some factors and a binary response variable. For instance, the response variable can describe the state of a patient health and one often assumes that it depends only on some part of factors. An important problem is to determine collections of significant factors. In this regard we turn to the MDR-method introduced by M.Ritchie and coauthors. Our recent paper provided the necessary and sufficient conditions for strong consistency of estimates of the prediction error employing the K-fold cross-validation and an arbitrary penalty function. Here we introduce the regularized versions of the mentioned estimates and prove for them the multidimensional CLT. Statistical variants of the CLT involving self-normalization are discussed as well. Keywords and phrases: binary response variable, significant factors, penalty function, cross-validation, MDR-method, SLLN for arrays, strong consistency, regularized estimates, multidimensional CLT, self-normalization.