Data Cleaning and Exploration with Machine Learning

Chapter 11: Decision Trees and Random Forest Classification

Decision trees and random forests are very popular classification models. This is partly because they are easy to train and interpret. They are also quite flexible. We can model complexity without necessarily having to increase the feature space or transform features. We do not even need to do anything special to apply the algorithm to multiclass problems, something we had to do with logistic regression.

On the other hand, decision trees can be less stable than other classification models, being fairly sensitive to small changes in the training data. Decision trees can also be biased when there is a significant class imbalance (when there are many more observations in one class than ...