diversityForest-package: Diversity Forests

In diversityForest: Innovative Complex Split Procedures in Random Forests Through Candidate Split Sampling

Diversity Forests

Description

The diversity forest algorithm is a split-finding approach that enables complex split procedures to be realized in random forest variants. This is achieved by drastically reducing the number of candidate splits evaluated at each node. It also avoids the variable selection bias seen in conventional random forests, where variables with many possible splits are selected too frequently for splitting (Strobl et al., 2007). For details, see Hornung (2022).

Details

This package currently features three methodologies which use the diversity forest algorithm:

• the basic form of diversity forests that uses univariable, binary splitting, which is also used in conventional random forests

• interaction forests (IFs) (Hornung & Boulesteix, 2022), which use bivariable splitting to model quantitative and qualitative interaction effects. IFs feature the Effect Importance Measure (EIM), which ranks the variable pairs with respect to the predictive importance of their quantitative and qualitative interaction effects. The individual variables can be ranked as well using EIM. For details, see Hornung & Boulesteix (2022).

• two variable importance measures (VIMs) designed for random forests for multi-class outcomes, the class-focused VIM and the discriminatory VIM. The class-focused VIM ranks variables based on their ability to distinguish individual outcome classes from all others, whereas the discriminatory VIM ranks variables based on their overall influence, irrespective of class relatedness.

Diversity forests with univariable, binary splitting can be constructed using the function divfor, interaction forests using the function interactionfor, and the class-focused and the discriminatory VIM can be computed using the function multifor. The former two methods support categorical, metric, and survival outcomes.

This package is a fork of the R package 'ranger' that implements random forests using an efficient C++ implementation. The documentation is in large parts taken from 'ranger', where some parts of the documentation may not apply to (the current version of) the 'diversityForest' package.

Details on further functionalities of the code that are not presented in the help pages of 'diversityForest' are found in the help pages of 'ranger', version 0.11.0, because 'diversityForest' is based on the latter version of 'ranger'. The code in the example sections can be used as a template for all basic application scenarios with respect to classification, regression and survival prediction.

References

• Hornung, R. (2022). Diversity forests: Using split sampling to enable innovative complex split procedures in random forests. SN Computer Science 3(2):1, <\Sexpr[results=rd]{tools:::Rd_expr_doi("10.1007/s42979-021-00920-1")}>.

• Hornung, R., Boulesteix, A.-L. (2022). Interaction forests: Identifying and exploiting interpretable quantitative and qualitative interaction effects. Computational Statistics & Data Analysis 171:107460, <\Sexpr[results=rd]{tools:::Rd_expr_doi("10.1016/j.csda.2022.107460")}>.

• Strobl, C., Boulesteix, A.-L., Zeileis, A., Hothorn, T. (2007). Bias in random forest variable importance measures: Illustrations, sources and a solution. BMC Bioinformatics 8:25, <\Sexpr[results=rd]{tools:::Rd_expr_doi("10.1186/1471-2105-8-25")}>.

• Wright, M. N., Ziegler, A. (2017). ranger: A fast Implementation of Random Forests for High Dimensional Data in C++ and R. Journal of Statistical Software 77:1-17, <\Sexpr[results=rd]{tools:::Rd_expr_doi("10.18637/jss.v077.i01")}>.

diversityForest documentation built on June 8, 2025, 1:23 p.m.