Random Competing Risks Forests for Large Data

TL;DR

This paper introduces largeRCRF, an R package for efficiently applying random competing risks forests to large datasets exceeding 100,000 samples, validated through simulations and real data analysis.

Contribution

The paper presents a new R package that enables scalable and accurate competing risks analysis on large datasets, filling a gap in existing tools.

Findings

The package produces results comparable to existing methods like randomForestSRC.

It significantly reduces computation time on large datasets.

Demonstrated applicability on a previously inaccessible large dataset.

Abstract

Random forests are a sensible non-parametric model to predict competing risk data according to some covariates. However, there are currently no packages that can adequately handle large datasets ($n > 100,000$). We introduce a new R package, largeRCRF, using the random competing risks forest theory developed by Ishwaran et al. (2014). We verify our package's validity and accuracy through simulation studies and show that its results are similar enough to randomForestSRC while taking less time to run. We also demonstrate the package on a large dataset that was previously inaccessible, using hardware requirements that are available to most researchers.