This paper discusses the hyperparameters and tuning strategies for the random forest (RF) algorithm. RF has several hyperparameters that need to be set by the user, such as the number of observations drawn for each tree, the number of variables considered for splits, the splitting rule, and the number of trees. While RF often performs well with default hyperparameters, tuning these parameters can improve performance. The paper reviews the literature on the impact of these hyperparameters on prediction performance and variable importance measures. It also presents a benchmark study comparing the performance and runtime of the tuneRanger R package, which automatically tunes RF using model-based optimization (MBO), with other tuning implementations in R and RF with default hyperparameters. The paper discusses the influence of various hyperparameters on RF performance, including the number of candidate variables (mtry), sampling scheme, node size, number of trees, and splitting rule. It also examines the influence of these hyperparameters on variable importance measures. The paper highlights the trade-off between stability and accuracy of individual trees and the importance of choosing appropriate hyperparameter values based on the dataset. It also discusses the use of out-of-bag observations for tuning and the impact of hyperparameters on the convergence of RF. The paper presents different tuning strategies, including grid search, random search, and sequential model-based optimization (SMBO). It describes the tuneRanger package, which uses SMBO to automatically tune RF hyperparameters. The package is based on the ranger and mlrMBO packages and provides a user-friendly interface for tuning RF. The paper also includes a benchmark study comparing the performance of tuneRanger with other tuning implementations on 39 datasets. The results show that tuneRanger outperforms the default RF and other tuning methods in terms of performance and runtime. The paper concludes that while RF is less tunable than other algorithms, tuning can still provide significant improvements in performance.This paper discusses the hyperparameters and tuning strategies for the random forest (RF) algorithm. RF has several hyperparameters that need to be set by the user, such as the number of observations drawn for each tree, the number of variables considered for splits, the splitting rule, and the number of trees. While RF often performs well with default hyperparameters, tuning these parameters can improve performance. The paper reviews the literature on the impact of these hyperparameters on prediction performance and variable importance measures. It also presents a benchmark study comparing the performance and runtime of the tuneRanger R package, which automatically tunes RF using model-based optimization (MBO), with other tuning implementations in R and RF with default hyperparameters. The paper discusses the influence of various hyperparameters on RF performance, including the number of candidate variables (mtry), sampling scheme, node size, number of trees, and splitting rule. It also examines the influence of these hyperparameters on variable importance measures. The paper highlights the trade-off between stability and accuracy of individual trees and the importance of choosing appropriate hyperparameter values based on the dataset. It also discusses the use of out-of-bag observations for tuning and the impact of hyperparameters on the convergence of RF. The paper presents different tuning strategies, including grid search, random search, and sequential model-based optimization (SMBO). It describes the tuneRanger package, which uses SMBO to automatically tune RF hyperparameters. The package is based on the ranger and mlrMBO packages and provides a user-friendly interface for tuning RF. The paper also includes a benchmark study comparing the performance of tuneRanger with other tuning implementations on 39 datasets. The results show that tuneRanger outperforms the default RF and other tuning methods in terms of performance and runtime. The paper concludes that while RF is less tunable than other algorithms, tuning can still provide significant improvements in performance.