How Many Trees in a Random Forest?

By Thais Mayumi Oshiro, Pedro Santoro Perez, José Augusto Baranauskas. 2012

911 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Information Systems

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Information Systems

Citations in the corpus

AutoScore: A Machine Learning–Based Automatic Clinical Score Generator and Its Application to Mortality Prediction Using Electronic Health Records (2020, methodology): “The parameters were selected according to the suggestions in previous literature [28,29], where ntree =100 and mtry = m 1/2 ( ntree : the number of trees grown; mtry : the number of variables randomly sampled as candidates at each split).”
Why do tree-based models still outperform deep learning on typical tabular data? (2022, background): “Furthermore, contrary to Hyperopt-sklearn, we do not tune the number of estimators for tree-based models, but set it to a constant high value (250 for random forest, following Oshiro et al. [2012], 1000 for gradient boosting models) and use early stopping for gradient boosting trees models.”
Random forest predictive modeling of mineral prospectivity with small number of prospects and data with missing values in Abra (Philippines) (2015, methodology): “Based on the study of Oshiro et al. (2012), the density of our datasets (with 10 predictors and training samples of 12 prospect and 12 non-prospect locations) is roughly dissimilar to that of Rodriquez-Galiano et al. (2014) and, therefore, we used a larger k of 20,000 to obtain stable predictions.”
Revisiting acoustic correlates of pharyngealization in Jordanian and Moroccan Arabic: Implications for formal representations (2017, methodology): “Hence, we implemented the same procedure as that suggested by Oshiro et al. (2012).”
Hyperparameters and tuning strategies for random forest (2018, background): “Using a large number of datasets, Oshiro et al. (2012) and Probst and Boulesteix (2017) show empirically that the biggest performance gain can often be achieved when growing the ﬁrst 100 trees.”
How to explain and predict the shape parameter of the generalized extreme value distribution of streamflow extremes using a big dataset (2018, background): “When increasing the number of trees hyperparameter, predictions become more accurate at the cost of increasing the computational time (Oshiro et al. 2012).”
A survey on semi-supervised learning (2019, background): “It has been empirically shown that, in supervised random forests, performance can substantially improve as the number of trees is increased (Oshiro et al. 2012).”
A non-model-based approach to bandwidth selection for kernel estimators of spatial intensity functions (2018, methodology): “…in which the quadrats are replaced by the cells in the pattern’s Delaunay tessellation, as shown in a 2007 University of Groningen PhD thesis by W. E. Schaap and in Schaap & van de Weygaert (2000), or the cells in the pattern’s Voronoi tessellation (Bernardeau & van de Weygaert, 1996; Ord, 1978).”
To tune or not to tune the number of trees in random forest? (2017, background): “Having this in mind, our observations are in agreement with the results of Oshiro et al. (2012), which conclude that after growing 128 trees no big gain in the AUC performance could be achieved by growing more trees.”