FRESA.CAD::ILAA: Estimating the exploratory residualization transform

Abstract

Multicollinearity among observed variables may have a large impact on statistical modeling and the discovery of associations between the observed variables and clinical outcomes. A viable method to address the multicollinearity is to find a suitable linear transform that mitigates the degree of collinearity. The Iterative Linear Association Analysis (ILAA) method was developed to explore the association among observed variables and to return a suitable linear transformation matrix based on variable residualization that effectively mitigates the degree of multicollinearity via controlling the maximum correlation measure present in the transformed dataset. This paper presents the software implementation of the ILAA method as an R function inside the FRESA.CAD 3.4.7 R package, hence providing researchers with a simple tool to explore tabular data in a new interpretable latent space.

Keywords

• Multicollinearity
• Latent variables
• Variable residualization
• Hyperdimensional spaces
• Decorrelation