Random Forest (RF) is a powerful classification tool. When given a set of data, RF generates a forest of decision trees, rather than a single tree. Each of these trees generates a prediction for a given set of attributes. The prediction from each H2O tree can be thought of as a vote; the most votes (or average vote) determines the prediction.
Number of features sampled for each split. If -1 is chosen, the square root of the number of features will be chosen.
Fewest allowed observations in a leaf
For numerical features, build a histogram of this many bins, then split at the best point. Low values lead too poorer splits, but higher values increase computation time.
For categorical features, group levels into this many bins, then split at the best point. Low values lead too poorer splits, but higher values increase computation time.