---
kind: concept
slug: random-forest
title: Random forest
summary: >-
  An ensemble of decision trees, each grown on a resampled training set, whose outputs are averaged. Insensitive to feature scaling on independent rows. Repeated measures need changed sampling or a mixed-effects hybrid. Covers subject-level bootstrap, historical summaries, mixed-effects hybrids, multivariate splits, regression forests, and plain forests in applied studies.
tags: [random-forest, tree-ensemble, longitudinal-data, mixed-effects, feature-scaling, regression, spam-filtering, fraud-detection, web-security, imbalanced-learning, predictive-maintenance, discrete-choice, parkinsons-disease, medical-diagnosis, remote-sensing, burned-area-mapping]
papers: ["1612.05890", "1904.08067", "2103.07678", "2112.15538", "2207.12877", "2208.04112", "2212.12343", "2405.11619", "2406.13547", "2607.19153", "2608.07471", "2609.07408", "2609.23687"]
related: [longitudinal-data-analysis, mixed-effects-model, ensemble-learning, stacked-generalization, feature-scaling, image-quality-assessment, text-classification, intrusion-detection, support-vector-machine, data-augmentation, generative-adversarial-network, discrete-choice-modeling, feature-selection, time-series-classification]
status: draft
updated_at: 2026-11-01
---

## Overview

A random forest averages many trees, and each tree is grown on a resampled
copy of the training data [arxiv:2208.04112 §2.1]. That design assumes each
row is an independent observation. With repeated measures, rows from one
subject are correlated. Then observation-level bootstrap samples contain rows
from nearly every subject, the trees become similar to each other, and the
out-of-bag (OOB) error looks better than it is, because OOB rows resemble
in-bag rows from the same subject [arxiv:2208.04112 §2.1] [arxiv:2208.04112 fig. 1].

Every variant below changes one of three things: how the training sample is
drawn, what the trees see as input (history summaries or random effects), or
what the split criterion optimises. The standard forest needs none of these
changes for independent rows, and its sensitivity to feature scaling is low
[arxiv:2212.12343 §5.1].

Thirteen cards on wikillm list this concept. The first is a review of longitudinal
extensions, which runs no experiments of its own, so its variants are reported
from the cited work and not measured here [arxiv:2208.04112 §5]. The second is
an empirical study of feature scaling, in which a random forest is one of 20
classifiers, tested on binary tabular datasets [arxiv:2212.12343 §4.1]. The
third uses forests as regressors for a single image-quality score, with
features that are not repeated measures [arxiv:1612.05890 §4.4]. The fourth
is a survey of text classification that lists forests among its classifiers
and reports no measured results for them [arxiv:1904.08067 §4.8]. The fifth
is an email spam and phishing classifier that compares a forest with linear
models on text features [arxiv:2405.11619 tab. 2]. The sixth is a card-fraud
comparison of seven standard classifiers, one of them a forest, on transaction
features [arxiv:2608.07471 fig. 4]. The seventh applies forests to ModSecurity
rule-firing features for SQL injection [arxiv:2406.13547 §3]. The eighth is a
predictive-maintenance study on rare machine failures, where a forest is the
classifier and also the objective that tunes the data generators
[arxiv:2607.19153 §3.3.2] [tab. 4]. The ninth is a discrete-choice paper, in
which a forest is one of several benchmarks for a neural choice model
[arxiv:2207.12877 tab. 1]. The tenth is a classifier comparison on spiral
drawings for Parkinson's disease, where a forest is one of nine default
classifiers [arxiv:2609.07408 p. 11]. The eleventh is a systematic review of machine
learning for disease diagnosis. It names forests among the methods of the
studies it reviews, but it trains no models and gives no forest results of its
own [arxiv:2112.15538 §6]. The twelfth is a review of machine learning for
mineral mapping from remote sensing. It describes forests as bagged trees that
vote or average, and lists them among the classifiers used in cited studies,
without results of its own [arxiv:2103.07678 §5.2.4] [tab. 2]. The thirteenth
is a dataset paper on burned-area mapping from satellite time series, in which
a forest is the only classifier and is trained on one row per grid cell and
year [arxiv:2609.23687 §2.2.15] [§2.2.16].

## Variants

### Standard forest (baseline)

Each tree is grown on a bootstrap sample of rows, and the trees are averaged.
This is the reference that the variants below modify. On 82 binary tabular
datasets with five-fold cross-validation, a forest's scores barely change when
the input features are rescaled. Friedman tests do not reject "no difference"
across the six scaling variants (all datasets: F1 p = 0.5169, G-Mean p = 0.4980)
[arxiv:2212.12343 tab. 4]. The paper reports decision trees and their ensembles
as nearly insensitive to the scaler [arxiv:2212.12343 §5.1] [arxiv:2212.12343 §5.3].
The paper configures the forest with 100 trees and the Gini criterion
[arxiv:2212.12343 tab. 1]. Its scores are for classification only, and
repeated measures are not tested [arxiv:2212.12343 §4.1].

### Subject-level resampling (RF++ and the two-stage bootstrap)

For clustered data without a time effect, the review sorts the fixes into
three groups. First is averaging, which collapses each subject's rows into
one before growing a standard forest. It loses within-subject variation and
was reported to degrade more than standard RF as the number of subjects
falls [arxiv:2208.04112 §2.2.1]. Second is the subject-level bootstrap, RF++:
resample subjects, and keep every row of each drawn subject in-bag
[arxiv:2208.04112 §2.2.2] [arxiv:2208.04112 fig. 2]. Third is the two-stage
bootstrap of Adler et al.: resample subjects, then keep one randomly chosen
row per drawn subject. In the simulations the review summarises, this scheme
predicted best against standard RF, averaging and RF++, but the review notes
that results depend on the setting [arxiv:2208.04112 §2.2.2].

Subject-level resampling only adjusts for clustering. There are no random
effects and the time order of observations is not used [arxiv:2208.04112 §2.2.2].

### Historical summaries for time-varying predictors

For a predictor measured over time, the historical RF first reduces each
subject's past to a summary. One example counts past values that fall below a
threshold inside a look-back window [arxiv:2208.04112 eq. 2]. Other summaries
are a relative frequency above a threshold and a mean, which the review says
give similar accuracy in a personal communication [arxiv:2208.04112 §2.3.1].
Only the frequency-based summaries keep a forest invariant to monotone
transforms of the predictor [arxiv:2208.04112 §2.3.1].

The window length and threshold are searched inside each split, which is
costly when there are many time-varying predictors. The htree authors search
on a random sub-sample instead [arxiv:2208.04112 §2.3.1]. The search grid and
sub-sample size are not stated. Variable importance is computed by deleting
each variable and refitting, so it can hide variables that are correlated
with others [arxiv:2208.04112 §2.3.1].

### Mixed-effects hybrids

The hybrids replace the fixed part of a linear mixed model with a tree or a
forest, and keep subject-specific random effects. The model is
$y_i = f(X_i) + Z_i b_i + \epsilon_i$ with normal random effects and errors
[arxiv:2208.04112 eq. 3]. MERT and MERF fit it by alternating a tree or forest
on the adjusted outcome with a mixed-model fit of the random effects
[arxiv:2208.04112 §2.3.2]. Observation-level resampling is kept in MERF, which
is valid only if the random effects absorb all within-subject correlation
[arxiv:2208.04112 §2.3.2].

RE-EM trees use the tree only to define leaves, then fit a mixed model with a
leaf indicator as fixed effect. Unlike MERT they allow a non-diagonal error
covariance, for example autocorrelated errors [arxiv:2208.04112 §2.3.2].
REEMforest averages several such trees. The stochastic variants add a
subject-specific Gaussian process [arxiv:2208.04112 §2.3.2].

For a binary outcome, GMERT, GMERF and the Bayesian BiMM forest extend the
same idea through a generalized mixed model [arxiv:2208.04112 §2.3.2]
[arxiv:2208.04112 §2.3.3]. GMET and GMERF initialise the linear predictor
with a plain GLM on all predictors, so they cannot run when there are more
predictors than observations [arxiv:2208.04112 §2.3.2] [arxiv:2208.04112 §5].

Prediction differs by case. For a new subject, use only the fixed part. For a
training subject, add its estimated random effect [arxiv:2208.04112 §2.3.2].

### Multivariate split criteria

When a subject's repeated outcomes are treated as one vector, the node
impurity changes. Segal's generalized sum of squares uses a node covariance
[arxiv:2208.04112 eq. 4]. Its covariance is estimated in the parent and reused
for both children, and all subjects must share observation times
[arxiv:2208.04112 §3.1.1]. The independence case reduces to a multivariate
sum of squares used in mvpart and in multivariate forests [arxiv:2208.04112 §3.1.1].
Distance-based criteria (db-MRT and related) need only a distance, so responses
can differ in length or timing. Prediction within a leaf then needs care,
because a plain average may not exist [arxiv:2208.04112 §3.1.1].
Likelihood-based criteria exist for multiple binary responses and for
multivariate normal responses [arxiv:2208.04112 §3.1.2].

These criteria apply directly only to time-invariant predictors. Time-varying
predictors need one of the history or mixed-effects techniques as well
[arxiv:2208.04112 §3].

### Regression forest with a log-determinant split objective

This variant is not a repeated-measures fix. The forests predict one quality
score per image from hand-built features, and the training targets are human
ratings [arxiv:1612.05890 §3]. The features are split into three groups, one
forest per group, and a linear layer combines the three outputs. The weights
are fitted by least squares [arxiv:1612.05890 §4.4] [eq. 7]. Each forest
prediction is the mean over its trees [eq. 6]. Node splits maximise a
log-determinant objective built on a conditional covariance from probabilistic
linear fitting [eq. 4] [eq. 5]. The
authors used T = 2000 trees per forest, chosen by sweeping 100 to 5000
[arxiv:1612.05890 §5.1]. Depth, leaf size and the feature-subset size are not
stated [arxiv:1612.05890 §4.4]. Author-reported Spearman correlation with human
scores is 0.931 under 5-fold cross-validation, against 0.921 for one forest on
concatenated features [arxiv:1612.05890 tab. 7] [tab. 6]. No code was checked
for this variant [arxiv:1612.05890 §5].

The paper does not say whether the forest outputs used to fit the linear layer
are out-of-fold [arxiv:1612.05890 eq. 7]. In-sample outputs would inflate that
fit, so use out-of-fold predictions when reimplementing.

## Applied uses of the standard forest

Seven of the cards use an unmodified forest as one classifier or baseline in a
comparison, and three surveys name forests without measuring them. The
cards do not record a repeated-measures design for any of these studies, and
each reported number comes from one author-run evaluation: a single split,
one cross-validation scheme, or repeated random draws from one dataset.

On email spam and phishing, the forest has 100 trees, each grown on a random
subset of features at every split, with a majority vote [arxiv:2405.11619 p. 9].
On a merged corpus of 82,486 emails, with one 80/20 split, the forest reaches
0.984 accuracy on TF-IDF features. A linear SVC reaches 0.991 on the same split
[arxiv:2405.11619 tab. 2]. With Word2Vec features the order reverses: the
forest scores 0.838 and the linear SVC 0.821 [arxiv:2405.11619 tab. 2]. Tree
depth, features per split and bootstrap settings are not stated
[arxiv:2405.11619 p. 9]. Other forest rows in the same table use different
subsets of the data and are not comparable with the proposed model
[arxiv:2405.11619 tab. 2].

On card fraud, the forest is one of seven candidates. The training split is
randomly undersampled to equal class counts, and Time and Amount are
standardized [arxiv:2608.07471 §3.4]. The forest's author-reported ROC AUC is
0.911, against 0.946 for logistic regression, the best single model, and 0.954
for a stacked ensemble [arxiv:2608.07471 fig. 4] [§4.2] [§5]. The paper uses one
Kaggle dataset, reports no run-to-run variance, and does not state the forest's
trees, depth or features per split [arxiv:2608.07471 §3.1] [§4.1]. The paper
also does not say whether model selection used the validation or the test split
[arxiv:2608.07471 fig. 3] [§4.2], so the AUC values may be optimistic.

On web-attack detection, each feature is a binary flag for one ModSecurity
SQL-injection rule that fired on a query [arxiv:2406.13547 §3]. The forest is
the top author-reported model at paranoia levels 3 and 4, with TPR at 1% FPR of
99.45, against 99.35 for L1 logistic regression at level 4
[arxiv:2406.13547 tab. 1]. That gap is 0.1 points on one balanced test set of
10,000 samples [arxiv:2406.13547 §4.1]. The 1% FPR is measured on that balanced
mix, so it is not a direct estimate of false positives in production
[arxiv:2406.13547 §4.1]. The paper says a non-linear model is harder to put
into ModSecurity than linear weights [arxiv:2406.13547 §3]. A forest has no
penalty term, so it cannot set rule weights to zero the way the L1 models do
[arxiv:2406.13547 §3] [arxiv:2406.13547 §4.4].

On rare machine failures, the forest is the classifier in every scenario. The
baseline gives it class weights balanced for the imbalance, with no
resampling [arxiv:2607.19153 §3.2.1]. Its author-reported PR-AUC is 0.835 with
precision 1.0 and recall 0.588 [arxiv:2607.19153 tab. 4]. Random undersampling
and SMOTE lower its PR-AUC to 0.724 and 0.7525 [arxiv:2607.19153 tab. 4]. The
best row adds synthetic failures from four WGAN-GP generators, one per failure
type, and reaches PR-AUC 0.857 and recall 0.882 at precision 0.570
[arxiv:2607.19153 tab. 4] [§3.2.5]. The forest's hyperparameters are
searched on a shallow grid that the paper does not list, and a deeper search
is said to have crashed the system [arxiv:2607.19153 §5.2]. The GAN
objective is the PR-AUC of a forest on the validation split, so the same
classifier that is compared across scenarios also sets the GAN hyperparameters
[arxiv:2607.19153 §3.3.2].
The paper runs one seed per configuration and does not test whether the 0.022
PR-AUC gap over the baseline is larger than run-to-run noise [arxiv:2607.19153 §5.1].
Compute is also large: about 14 hours for the multi-GAN scenario
[arxiv:2607.19153 §5.1].

In the discrete-choice paper the forest is a benchmark with a grid of 50 to
400 trees and depths of 5 to 20, chosen on validation loss [arxiv:2207.12877 §6.2].
On Swissmetro its log-likelihood loss is the lowest in the table, 0.523, and
its accuracy is 0.776, below the 0.793 of the proposed network [arxiv:2207.12877 tab. 1].
On Expedia its loss is 2.712, worse than the 2.482 of multinomial logit
[arxiv:2207.12877 tab. 1]. The authors tie the forest's overfitting on Expedia
to its larger input dimension [arxiv:2207.12877 §6.3] [§6.4]. The paper also
says training the proposed network takes about an order of magnitude longer
than a forest, and that it is unclear whether this is inherent
[arxiv:2207.12877 app. E]. All of these numbers are author-reported on one 10-fold scheme.

On spiral drawings for Parkinson's disease, the forest is one of nine classifiers
with library default settings and no tuned values [arxiv:2609.07408 p. 11]. Its
inputs are eleven statistics per image [arxiv:2609.07408 p. 9]. The paper
reports no held-out test set, and every score comes from stratified 10-fold
cross-validation on 72 training images, repeated ten times
[arxiv:2609.07408 p. 8] [p. 11]. The forest's AUC is 0.90 on the plain drawings
and 0.97 on the augmented set, and the edge-preprocessed sets give 0.85 and
0.89 [arxiv:2609.07408 fig. 12] [fig. 13]. The card does not record how
augmented copies of one image are assigned to folds, so the 0.97 may be
optimistic. Check that before relying on it.

On burned-area mapping, the forest is the only classifier. Each row is one
500 m grid cell in one year, built from daily values of the four BGR-NIR bands
[arxiv:2609.23687 §2.2.15]. The grid search covers 100, 200 and 300
trees and maximum depths of 3, 5 and 7 [arxiv:2609.23687 §2.2.16]. Each of the
three relabelling approaches is repeated 200 times with a new random draw, and
30% of the samples are held out [arxiv:2609.23687 §2.2.16] [§2.2.17]. The
author-reported mean test F1 for the binary labels is 0.92 to 0.95
[arxiv:2609.23687 tab. 7]. Non-burned cells are 86% to 88% of the grid
[arxiv:2609.23687 §3.3], so overall IoU stays high even when burned-area IoU is
low. In the best validation setup, burned-area IoU is 0.49 to 0.71
[arxiv:2609.23687 tab. 8]. The grid is spatially adjacent and the paper does not
discuss spatial autocorrelation, so a random cell split may be optimistic
[arxiv:2609.23687 §2.2.9].

The text survey names forests as one of its classifiers and describes them
in its section on classifiers, with no measured results
[arxiv:1904.08067 §4.8]. Use it for the vocabulary, not for a ranking. The
mineral-mapping review does the same for remote sensing. It describes the forest
as suited to large, complicated datasets without a parametric form
[arxiv:2103.07678 §5.2.4] and reports no forest result. Its two numeric
findings come from other studies and do not involve forests
[arxiv:2103.07678 §5].

The disease-diagnosis review lists a forest alongside Gaussian and Bernoulli
naive Bayes for coronary heart disease, citing one study [arxiv:2112.15538 §6]. It gives
no accuracy or other metric for that forest, and it reports the reviewed
accuracies only as the authors' own [arxiv:2112.15538 §7]. Use it as a pointer
to the primary studies, not as evidence for the forest's performance.

## When to use which

| Situation | Use | Why |
|---|---|---|
| Tabular features, rows are independent | Standard forest | The clustering problems below arise only with repeated measures [arxiv:2208.04112 §2.1] |
| Forest is the final model and features have mixed scales | Standard forest, no scaler needed for the forest itself | No significant scaler effect for RF on 82 binary datasets [arxiv:2212.12343 tab. 4]. Scaling still matters for any scale-sensitive model in the same pipeline [arxiv:2212.12343 §5.1] |
| Base learner is not a tree (for example Perceptron bagging) | Expect scaler sensitivity; scale the inputs | Bagging of Perceptrons rises from F1 0.39 unscaled to 0.56 with the standard scaler [arxiv:2212.12343 tab. 3] |
| Repeated measures, no time effect needed, simplest change | Subject-level bootstrap (RF++) or the two-stage bootstrap | Only sampling changes. Compute OOB per subject. Two-stage was best in the cited simulations [arxiv:2208.04112 §2.2.2] |
| One row per subject is acceptable, and the subject count is large | Averaging, with caution | Reported to degrade more than standard RF as the number of subjects falls [arxiv:2208.04112 §2.2.1] |
| Predictors change over time and history matters | Historical RF | Summaries capture the past. Cost rises with the number of time-varying predictors [arxiv:2208.04112 §2.3.1] |
| Subject-specific offsets should be explicit, and the outcome is continuous | MERT, MERF or RE-EM hybrids | Reported to beat a linear mixed model and standard RF on high-dimensional data in one cited study [arxiv:2208.04112 §2.3.2] |
| Binary outcome, more predictors than observations | Avoid GMET and GMERF | Their GLM start fails when predictors outnumber observations [arxiv:2208.04112 §2.3.2] |
| Several outcomes per subject, time-invariant predictors | Multivariate criteria | Impurity is taken over the whole response vector [arxiv:2208.04112 §3.1] |
| One scalar regression target from several groups of hand-built features | One forest per group, combined by a linear layer | Author-reported 0.931 Spearman under 5-fold CV, against 0.919 for the best baseline in that table [arxiv:1612.05890 tab. 7]. Only one cited study, so no general claim is made |
| Email text with TF-IDF features, a classical model is wanted | Linear SVC first; a forest is a second choice | Author-reported 0.991 for SVC against 0.984 for the forest on one split [arxiv:2405.11619 tab. 2]. Only one split was run |
| Email text with Word2Vec features | Forest over linear SVC | Author-reported 0.838 for the forest against 0.821 for SVC on the same split [arxiv:2405.11619 tab. 2]. The ordering is feature-dependent, so do not carry it to other features |
| Rare-positive tabular labels (fraud), one best model is wanted from a comparison | Logistic regression before a forest | Author-reported AUC 0.946 for logistic regression against 0.911 for the forest, on one undersampled split [arxiv:2608.07471 §4.2]. The gap is small and no variance is reported, so treat the order as a single run |
| Binary rule-firing features, rules should be dropped or weighted sparsely | L1-regularised linear model | Only the linear models have an L1 penalty that sets weights to zero [arxiv:2406.13547 §3] [arxiv:2406.13547 §4.4] |
| Binary rule-firing features, accuracy is the priority and per-query rule runs are affordable | Forest or L1 logistic regression; the gap is small | Author-reported TPR at 1% FPR of 99.45 for the forest against 99.35 for L1 LR at PL4 [arxiv:2406.13547 tab. 1]. The paper finds linear weights easier to deploy in ModSecurity [arxiv:2406.13547 §3] |
| Rare positives in tabular data, recall must rise, and GAN training time is affordable | Forest over GAN-augmented data, one generator per failure type | Author-reported PR-AUC 0.857 against 0.835 for the cost-sensitive forest, and recall 0.882 against 0.588 [arxiv:2607.19153 tab. 4]. One run per configuration, so the gap is not tested for noise [arxiv:2607.19153 §5.1] |
| Rare positives in tabular data, GAN training is not affordable | Cost-sensitive forest (class weights balanced) | In the same table it beats SMOTE and random undersampling on PR-AUC: 0.835 against 0.7525 and 0.724 [arxiv:2607.19153 tab. 4]. Precision is high (1.0) but recall is low (0.588) [tab. 4] |
| Choice prediction where each choice set has a different size, forest is a baseline | Pad choice sets with dummy items, or use a model that masks unavailable items | Forests need fixed-size inputs. The paper pads with dummy products whose features are zero except a large log-price [arxiv:2207.12877 app. E] |
| Labelled per-cell time series of satellite bands, binary burned vs not burned | Standard forest as a baseline, scored on burned-area IoU as well as overall IoU | Author-reported F1 0.92 to 0.95 for binary labels, but NB is 86% to 88% of cells, and burned-area IoU is 0.49 to 0.71 in the best setup [arxiv:2609.23687 tab. 7] [tab. 8] [§3.3] |
| Small image-derived feature set, library defaults, small model and fast prediction | Standard forest | Author-reported AUC 0.90 on plain drawings, against 0.85 after edge preprocessing [arxiv:2609.07408 fig. 12] [fig. 13]. Model size is about 61 KB and fixed across the four datasets [arxiv:2609.07408 fig. 40]. One small dataset, so no general claim |

## Implementation sketch

The shared core is a per-tree training sample. Clustered data needs a
subject-level draw, and the forest uses each drawn subject's rows as the
review describes [arxiv:2208.04112 §2.2.2] [arxiv:2208.04112 fig. 2].

```pseudocode
# Per-tree training sample for clustered data  [arxiv:2208.04112 §2.2.2]
def tree_sample(rows_by_subject, two_stage):
    subjects = list(rows_by_subject.keys())
    # Draw subjects with replacement, not rows  [arxiv:2208.04112 §2.2.2]
    drawn = sample_with_replacement(subjects, size=len(subjects))
    in_bag = []
    for subject in drawn:
        rows = rows_by_subject[subject]
        if two_stage:
            # Adler et al.: one random row per drawn subject  [§2.2.2]
            in_bag.append(random_choice(rows))
        else:
            # RF++: every row of the drawn subject is in-bag  [§2.2.2]
            in_bag.extend(rows)
    # Whole subjects that were never drawn are out-of-bag  [§2.2.2]
    out_of_bag = [
        row for subject in subjects if subject not in drawn
        for row in rows_by_subject[subject]
    ]
    return in_bag, out_of_bag
```

The review does not spell out the OOB rule above; it follows from subject-level
resampling [arxiv:2208.04112 §2.2.2]. Per-subject OOB is the estimate to use
[arxiv:2208.04112 §2.2.2].

For independent rows the sample is a plain row bootstrap, and no scaling step
is needed for the forest itself [arxiv:2212.12343 §5.1]. Scalers are fitted on
the training part of each fold only, as the scaling study does
[arxiv:2212.12343 §4.2].

## Contradictions

The spam paper ranks a linear SVC above the forest on TF-IDF features, while the
ModSecurity paper ranks the forest above the linear models at paranoia levels
3 and 4 [arxiv:2405.11619 tab. 2] [arxiv:2406.13547 tab. 1]. The fraud paper
ranks logistic regression above the forest on AUC [arxiv:2608.07471 fig. 4].
The tasks, features and test sets differ, so these orderings are reported side
by side and not resolved. The spam paper also ranks the forest above the SVC on
Word2Vec features [arxiv:2405.11619 tab. 2].

The disease-diagnosis review's text and its tables credit the heart-disease
methods to different studies. The text credits Sun et al. with a forest among
the heart-disease methods [arxiv:2112.15538 §6], while its table credits
Bemando et al. [arxiv:2112.15538 tab. 3]. The card does not resolve which
attribution is right, so this page does not use either one.

In the discrete-choice paper the forest has the lowest Swissmetro loss, while
the proposed network has the higher accuracy on the same split. The two metrics
point to different winners, and the paper does not settle which one matters
[arxiv:2207.12877 tab. 1]. On Expedia the same table ranks the forest below
multinomial logit on loss, so the forest's rank changes with the dataset.

The predictive-maintenance paper's text and table disagree on the forest's F1.
Its §4.1 calls the multi-GAN forest F1 (0.6923) the highest across conditions,
while the baseline forest F1 in tab. 4 is 0.740, and the same section later
says the multi-GAN F1 is slightly lower [arxiv:2607.19153 §4.1] [tab. 4].
Its text and figures also differ on the RUS forest precision (0.2156 against
0.216) [arxiv:2607.19153 §4.1] [fig. 10]. Use the table values
[arxiv:2607.19153 tab. 4]. The paper's two rows labelled "Wide RF grid" repeat
values close to the SVM rows, which looks like a copy error, so they are not
used here [arxiv:2607.19153 tab. 4].

Within the longitudinal review, the cited simulations rank the two-stage
bootstrap first, but the review says results depend on the setting
[arxiv:2208.04112 §2.2.2]. The scaling study and the review do not overlap
[arxiv:2212.12343 §4.1].

## Papers

- [Learning a No-Reference Quality Metric for Single-Image Super-Resolution](/papers/1612.05890): uses three random forests, one per feature group, whose outputs are linearly combined to score super-resolved images.
- [A review of machine learning in processing remote sensing data for mineral exploration](/papers/2103.07678): surveys ML for mineral mapping from satellite and airborne data, describes random forests among its classifiers, and runs no experiments.
- [A multi-temporal dataset for mapping burned areas in the Brazilian Cerrado using time series of remote sensing imagery](/papers/2609.23687): builds a labelled WFI cell dataset and uses a random forest as its only classifier; reports binary F1 of 0.92 to 0.95 and burned-area IoU of 0.49 to 0.71 in the best setup.
- [Text Classification Algorithms: A Survey](/papers/1904.08067): maps the text-classification pipeline and lists random forests among its classifiers, with no experiments.
- [Machine learning based disease diagnosis: A comprehensive review](/papers/2112.15538): a systematic review of ML for diagnosis that names forests among the reviewed methods, with no experiments of its own.
- [Longitudinal random-forest review](/papers/2208.04112): maps CART-based forest extensions for repeated measures, with equations and a software table, and runs no experiments.
- [The choice of scaling technique matters for classification performance](/papers/2212.12343): shows that a random forest is nearly insensitive to the choice of feature scaler across 82 binary datasets, while scale-sensitive classifiers react strongly.
- [Novel Interpretable and Robust Web-based AI Platform for Phishing Email Detection](/papers/2405.11619): compares a 100-tree forest with linear SVC and Naive Bayes on TF-IDF and Word2Vec email features, and reports the linear SVC as best.
- [Application of Artificial Intelligence for Fraudulent Banking Operations Recognition](/papers/2608.07471): compares seven standard classifiers, including a 0.911-AUC forest, on undersampled card transactions, and reports a stacked ensemble as best.
- [ModSec-Learn: Boosting ModSecurity with Machine Learning](/papers/2406.13547): learns weights over ModSecurity SQL-injection rules with a linear SVM, logistic regression or a random forest, and reports the forest as best at paranoia levels 3 and 4.
- [Breaking the Homogeneity Assumption: Specialized Multi-Generator Adversarial Learning for Rare Failure Detection in Predictive Maintenance](/papers/2607.19153): trains one WGAN-GP per failure type and pools the synthetic failures; a random forest is the classifier and reports PR-AUC 0.857 against 0.835 for the cost-sensitive baseline, on one run.
- [Representing Random Utility Choice Models with Neural Networks](/papers/2207.12877): benchmarks a tuned random forest against its neural choice model; the forest has the lowest Swissmetro loss, and the proposed model has the higher accuracy on both datasets.
- [Impact of canny edge detection preprocessing on performance of machine learning models for Parkinson's disease classification](/papers/2609.07408): trains nine default classifiers, including a forest, on spiral drawings with and without edge preprocessing and augmentation; the forest's AUC falls after edge preprocessing and reaches 0.97 on the augmented plain set.