Which hyperparameter specifies how many samples are required to split a decision node?

The hyperparameter that specifies how many samples are required to split a decision node is min_samples_split. This parameter plays a crucial role in determining whether a node should be split into further child nodes based on the number of samples it contains. If the number of samples in a node is less than the specified value for min_samples_split, the node will not be split, and it will become a leaf node. This helps with controlling the complexity of the decision tree by preventing overfitting, as it ensures that nodes with insufficient data do not get split needlessly.

The other options represent different aspects of decision tree configurations but do not specifically define the sample requirement for splits. For example, min_samples_leaf defines the minimum number of samples required to be at a leaf node, which is different from the splitting criteria. Max_depth restricts how deep the decision tree can grow, thus influencing the overall size but not directly tied to splitting criteria based on sample size. The splitter refers to the strategy used to choose the split at each node (like 'best' or 'random') but does not define a numeric requirement concerning the samples in a node.