CHAID Trees can have multiple nodes (more than 2), so decision trees are not always binary.

There are many different tree building algorithms and the Random Forest algorithm actually creates an ensemble of decision trees. In the original paper, the authors used a slight variation of the CART algorithm. However its not necessary to use CART to build a Random Forest model; other researchers have extended the concept by using tree learners built using different algorithms to generate a Random Forest model.

It is also not necessary that the independent variables which are used to build a decision tree necessarily have to be binary valued (0-1 valued or one-hot encoded categorical variables). Doing so would have severely restricted the usability of these algorithms. Each of the nodes may contain multiple categories after the split. For example, a variable with weather category as windy, foggy, rainy and sunny may be split into two nodes - Node 1 : windy + sunny, and Node 2 : foggy and rainy.

To answer your specific query: both scikit-learn and R use CART for building random forest models, so their base learners are binary trees.

CHAID Trees can have multiple nodes (more than 2), so decision trees are not always binary.

There are many different tree building algorithms and the Random Forest algorithm actually creates an ensemble of decision trees. In the original paper, the authors used a slight variation of the CART algorithm. However its not necessary to use CART to build a Random Forest model; other researchers have extended the concept by using tree learners built using different algorithms to generate a Random Forest model.

It is also not necessary that the independent variables which are used to build a decision tree necessarily have to be binary valued (0-1 valued or one-hot encoded categorical variables). Doing so would have severely restricted the usability of these algorithms. Each of the nodes may contain multiple categories after the split. For example, a variable with weather category as windy, foggy, rainy and sunny may be split into two nodes - Node 1 : windy + sunny, and Node 2 : foggy and rainy.

To answer your specific query: both scikit-learn and R use CART for building random forest models, so their base learners are binary trees.

CHAID Trees can have multiple nodes (more than 2), so decision trees are not always binary.

There are many different tree building algorithms and the Random Forest algorithm actually creates an ensemble of decision trees. In the original paper, the authors used a slight variation of the CART algorithm. However its not necessary to use CART to build a Random Forest model; other researchers have extended the concept by using tree learners built upon different algorithms to generate a Random Forest model.

It is also not necessary that the independent variables which are used to build a decision tree necessarily have to be binary valued (0-1 valued or one-hot encoded categorical variables). Doing so would have severely restrict the usability of these algorithms.

To reply to your specific query: scikit-learn and R both use CART for building random forest models, so the base learners are binary trees.

CHAID Trees can have multiple nodes (more than 2), so decision trees are not always binary.

There are many different tree building algorithms and the Random Forest algorithm actually creates an ensemble of decision trees. In the original paper, the authors used a slight variation of the CART algorithm. However its not necessary to use CART to build a Random Forest model; other researchers have extended the concept by using tree learners built using different algorithms to generate a Random Forest model.

It is also not necessary that the independent variables which are used to build a decision tree necessarily have to be binary valued (0-1 valued or one-hot encoded categorical variables). Doing so would have severely restricted the usability of these algorithms. Each of the nodes may contain multiple categories after the split. For example, a variable with weather category as windy, foggy, rainy and sunny may be split into two nodes - Node 1 : windy + sunny, and Node 2 : foggy and rainy.

To answer your specific query: both scikit-learn and R use CART for building random forest models, so their base learners are binary trees.