Question

When we have K>2 classes, another possibility is to train K trees, each separating one class from all others. Discuss how such trees are trained and how the final decision can be made.

Answer 1

These trees are built by splitting data repeatedly into small samples and making predictions in machine learning. The data to be split is training data.

Such trees can be trained by transferring data to leaves from the root node. Using predictor variables the data is split and it makes child nodes homogeneous. The leaves, as the data is passed down, contain only a single class as a result. For the categorical outcomes, cross-entropy of the split can be considered as well.