Dive into Decision Trees and Forests: A Theoretical Demonstration

Abstract: Based on decision trees, many fields have arguably made tremendous progress in recent years. In simple words, decision trees use the strategy of "divide-and-conquer" to divide the complex problem on the dependency between input features and labels into smaller ones. While decision trees have a long history, recent advances have greatly improved their performance in computational advertising, recommender system, information retrieval, etc. We introduce common tree-based models (e.g., Bayesian CART, Bayesian regression splines) and training techniques (e.g., mixed integer programming, alternating optimization, gradient descent). Along the way, we highlight probabilistic characteristics of tree-based models and explain their practical and theoretical benefits. Except machine learning and data mining, we try to show theoretical advances on tree-based models from other fields such as statistics and operation research. We list the reproducible resource at the end of each method.