Classification Trees

This chapter investigates decision trees, one of the most popular supervised learning algorithms. In addition to the Summary, there are six sections in this chapter. Section 9.1 describes classification and regression trees (CARTs), including decision tree structure, predictor space and its regions, and the steps of building a decision tree. Section 9.2 describes the algorithms for building a decision tree, including the greedy algorithm, optimization criterion (Gini impurity and cross-entropy), and tree pruning (cost complexity pruning and information gain pruning). Section 9.3 introduces the framework of tidymodels, including data sampling with rsample, preprocessing with recipes, model specification with parsnip (training models and making predictions), modeling workflows with workflows, and model validation with yardstick. Section 9.4 illustrates how to create classification trees using the rpart package with tidymodels. Section 9.5 provides some remarks on classification trees, including comparisons of CART, C4.5, and C5.0, and the advantages and disadvantages of CART. Section 9.6 provides remarks on tidymodels, including comparing machine learning within the framework of tidymodels and with R&s;s functions, as well as with the Caret package.