On the overestimation of random forest’s out-of-bag error

Abstract

The ensemble method random forests has become a popular classification tool in bioinformatics and related fields. The out-of-bag error is an error estimation technique often used to evaluate the accuracy of a random forest and to select appropriate values for tuning parameters, such as the number of candidate predictors that are randomly drawn for a split, referred to as mtry. However, for binary classification problems with metric predictors it has been shown that the out-of-bag error can overestimate the true prediction error depending on the choices of random forests parameters. Based on simulated and real data this paper aims to identify settings for which this overestimation is likely. It is, moreover, questionable whether the out-of-bag error can be used in classification tasks for selecting tuning parameters like mtry, because the overestimation is seen to depend on the parameter mtry. The simulation-based and real-data based studies with metric predictor variables performed in this paper show that the overestimation is largest in balanced settings and in settings with few observations, a large number of predictor variables, small correlations between predictors and weak effects. There was hardly any impact of the overestimation on tuning parameter selection. However, although the prediction performance of random forests was not substantially affected when using the out-of-bag error for tuning parameter selection in the present studies, one cannot be sure that this applies to all future data. For settings with metric predictor variables it is therefore strongly recommended to use stratified subsampling with sampling fractions that are proportional to the class sizes for both tuning parameter selection and error estimation in random forests. This yielded less biased estimates of the true prediction error. In unbalanced settings, in which there is a strong interest in predicting observations from the smaller classes well, sampling the same number of observations from each class is a promising alternative.

Reconstruction of Multidecadal Country-Aggregated Hydro Power Generation in Europe Based on a Random Forest Model

“…Although the two-step random forest model showed good performance, especially for Spain (Pearson's correlation coefficient of 0.90 for reservoir-based and 0.98 for run-of-river generation from 5-fold cross-validation), this validation method usually gives optimistic results [16]. One possible explanation is that despite being randomly permuted, the validation dataset in each cross-validation is still an average of multiple random samples from the same dataset, and thus the prediction is not totally independent of the training set.…”

Deriving Accurate Lipid Classification based on Molecular Formula

“…To evaluate the accuracy for the Random Forest models, we used the out-of-bag training accuracy. While not strictly equivalent to explicit cross-validation, the accuracy metric provided by out-of-bag training accuracy is a sufficient proxy, even though for unbalanced datasets such as ours, the out-of-bag training accuracy underestimates the error rate (Janitza and Hornung, 2018); however, from our own experience, this underestimation is not significant. For precision, the classic definition of precision can be applied for sets of inputs on known label and their model-generated labeling.…”

Deriving Lipid Classification Based on Molecular Formulas

“…While not strictly equivalent to explicit cross-validation, the accuracy metric provided by out-of-bag training accuracy is a sufficient proxy. In fact, for unbalanced datasets such as ours, the out-of-bag training accuracy can underestimate the error rate [63]. Therefore, the use of k-fold cross-validation with Random Forest would only reduce training sample size and could substantially overestimate the error.…”