Prediction modelling with many correlated and zero-inflated predictors

This repository contains the R code to reproduce the results in the paper titled "Prediction modelling with many correlated and zero-inflated predictors: assessing a nonnegative garrote approach", which is accessible as a preprint version on arXiv. Access it here.

Overview

Simulation study to examine the performance of the five regularized regression approach in terms of predictive accuracy and predictor selection in the presence of many correlated and zero-inflated predictors. The objective was to identify an approach with good predictive accuracy and the ability to select a parsimonious set of predictors.

The following methods are implemented:

• ridge regression
• lasso regression
• ridge-garrote approach
• ridge-lasso approach
• lasso-ridge approach

Contents

This repository is organized in the following manner:

• src: contains the code and R functions for the simulation study
• auxiliary: contains supplementary R Markdown reports focusing on specific aspects of the simulation study
• example: contains the code for a real-life case study on predicting kidney function using peptidomic features (data not provided)

Installation

You can install the development version from GitHub:

# install.packages("devtools")
devtools::install_github("mgregorich/ZIPSel")

Prerequisites

The code uses the statistical software R (>= 4.2.3)

References

Breiman, L. (1995). "Better subset regression using the nonnegative garrote." Technometrics 37(4): 373-384.