Hello Everyone,
I am know that we should select k features for each decision tree in random forest,
my question is how to how should select those features like with replacement or without replacement method.
To reduce overfitting of your random forest, I would go first for a model without replacement. In this way we can ensure that not a single feature make a bias in the decision-making process in the roots (individual trees).
Then you can use cross-validation to see how well it performs. Please note that this is not a general rule (doing without replacement).