Multivariate Random Forest with Compositional Responses

Non linear regression with compositional responses and Euclidean predictors is performed. The compositional data are first transformed using the additive log-ratio transformation, and then the multivariate random forest of Rahman R., Otridge J. and Pal R. (2017), , is applied.

Random Forest with Canonical Correlation Analysis

Random Forest with Canonical Correlation Analysis (RFCCA) is a random forest method for estimating the canonical correlations between two sets of variables depending on the subject-related covariates. The trees are built with a splitting rule specifically designed to partition the data to maximize the canonical correlation heterogeneity between child nodes. The method is described in Alakus et al. (2021) . 'RFCCA' uses 'randomForestSRC' package (Ishwaran and Kogalur, 2020) by freezing at the version 2.9.3. The custom splitting rule feature is utilised to apply the proposed splitting rule. The 'randomForestSRC' package implements 'OpenMP' by default, contingent upon the support provided by the target architecture and operating system. In this package, 'LAPACK' and 'BLAS' libraries are used for matrix decompositions.

Variable Selection using Random Forests

Variable selection from random forests using both backwards variable elimination (for the selection of small sets of non-redundant variables) and selection based on the importance spectrum (somewhat similar to scree plots; for the selection of large, potentially highly-correlated variables). Main applications in high-dimensional data (e.g., microarray data, and other genomics and proteomics applications).

Unbiased Variable Importance for Random Forests

Computes a novel variable importance for random forests: Impurity reduction importance scores for out-of-bag (OOB) data complementing the existing inbag Gini importance, see also . The Gini impurities for inbag and OOB data are combined in three different ways, after which the information gain is computed at each split. This gain is aggregated for each split variable in a tree and averaged across trees.

Easy Spatial Modeling with Random Forest

Automatic generation and selection of spatial predictors for spatial regression with Random Forest. Spatial predictors are surrogates of variables driving the spatial structure of a response variable. The package offers two methods to generate spatial predictors from a distance matrix among training cases: 1) Moran's Eigenvector Maps (MEMs; Dray, Legendre, and Peres-Neto 2006 ): computed as the eigenvectors of a weighted matrix of distances; 2) RFsp (Hengl et al. ): columns of the distance matrix used as spatial predictors. Spatial predictors help minimize the spatial autocorrelation of the model residuals and facilitate an honest assessment of the importance scores of the non-spatial predictors. Additionally, functions to reduce multicollinearity, identify relevant variable interactions, tune random forest hyperparameters, assess model transferability via spatial cross-validation, and explore model results via partial dependence curves and interaction surfaces are included in the package. The modelling functions are built around the highly efficient 'ranger' package (Wright and Ziegler 2017 ).

Prediction Intervals with Random Forests and Boosted Forests

Implements various prediction interval methods with random forests and boosted forests. The package has two main functions: pibf() produces prediction intervals with boosted forests (PIBF) as described in Alakus et al. (2022) and rfpi() builds 15 distinct variations of prediction intervals with random forests (RFPI) proposed by Roy and Larocque (2020) .

Binomial Random Forest Feature Selection

The 'binomialRF' is a new feature selection technique for decision trees that aims at providing an alternative approach to identify significant feature subsets using binomial distributional assumptions (Rachid Zaim, S., et al. (2019)) . Treating each splitting variable selection as a set of exchangeable correlated Bernoulli trials, 'binomialRF' then tests whether a feature is selected more often than by random chance.

Multiple Imputation using Chained Random Forests

An R package for multiple imputation using chained random forests. Implemented methods can handle missing data in mixed types of variables by using prediction-based or node-based conditional distributions constructed using random forests. For prediction-based imputation, the method based on the empirical distribution of out-of-bag prediction errors of random forests and the method based on normality assumption for prediction errors of random forests are provided for imputing continuous variables. And the method based on predicted probabilities is provided for imputing categorical variables. For node-based imputation, the method based on the conditional distribution formed by the predicting nodes of random forests, and the method based on proximity measures of random forests are provided. More details of the statistical methods can be found in Hong et al. (2020) .

Causal Effect Random Forest of Interaction Tress

Fits a Causal Effect Random Forest of Interaction Tress (CERFIT) which is a modification of the Random Forest algorithm where each split is chosen to maximize subgroup treatment heterogeneity. Doing this allows it to estimate the individualized treatment effect for each observation in either randomized controlled trial (RCT) or observational data. For more information see X. Su, A. T. Peña, L. Liu, and R. A. Levine (2018) .

Oblique Decision Random Forest for Classification and Regression

The oblique decision tree (ODT) uses linear combinations of predictors as partitioning variables in a decision tree. Oblique Decision Random Forest (ODRF) is an ensemble of multiple ODTs generated by feature bagging. Both can be used for classification and regression as supplements to the classical CART of Breiman (1984) and Random Forest of Breiman (2001) respectively.