You may want to look into extremely randomized trees, which have a slightly different implementation but can sometimes perform better than random forests.
In ensemble methods, each model learns differently in terms of the subset of the dataset and the subset of the feature vector used for training. These subsets are taken randomly. Extremely randomized trees possess a high randomness factor in the way they compute the splits and the subset of the features selected. Unlike random forests, in which the splitting threshold is chosen randomly, in extremely randomized trees, a discriminative threshold is used as the splitting rule. Due to this, the overall variance of the ensemble decreases and the overall performance may be better.
The scikit-learn implementation of extremely randomized trees can be found at the following link: https://bit.ly/2zWsNNS. H2O also supports extremely randomized trees.