Boosting is a general method for improving the accuracy of any learning algorithm. This chapter focuses on AdaBoost, a popular boosting algorithm, and reviews recent research on boosting, including analyses of AdaBoost's training and generalization errors, its connection to game theory and linear programming, its relationship with logistic regression, extensions for multiclass classification, methods to incorporate human knowledge, and experimental applications. Machine learning involves automatic techniques for making accurate predictions based on past observations. For example, building an email filter to distinguish spam from non-spam involves gathering examples, labeling them, and using a machine-learning algorithm to generate a classification rule. However, creating a highly accurate rule is challenging, while rough rules can be easier to create. Boosting leverages this by combining multiple weak rules into a more accurate one. Boosting starts with a base learning algorithm to generate weak rules. Each round, the algorithm is given a different distribution of training examples, with more weight on examples misclassified in previous rounds. This forces the base learner to focus on difficult cases. The weak rules are then combined, typically through a weighted majority vote. The chapter discusses the choice of base learning algorithm, but leaves it open to allow a general boosting procedure. Boosting is a provably effective method for combining rough rules into a highly accurate prediction rule. The chapter provides an overview of recent work on boosting, with a focus on AdaBoost, which has been extensively studied theoretically and empirically.Boosting is a general method for improving the accuracy of any learning algorithm. This chapter focuses on AdaBoost, a popular boosting algorithm, and reviews recent research on boosting, including analyses of AdaBoost's training and generalization errors, its connection to game theory and linear programming, its relationship with logistic regression, extensions for multiclass classification, methods to incorporate human knowledge, and experimental applications. Machine learning involves automatic techniques for making accurate predictions based on past observations. For example, building an email filter to distinguish spam from non-spam involves gathering examples, labeling them, and using a machine-learning algorithm to generate a classification rule. However, creating a highly accurate rule is challenging, while rough rules can be easier to create. Boosting leverages this by combining multiple weak rules into a more accurate one. Boosting starts with a base learning algorithm to generate weak rules. Each round, the algorithm is given a different distribution of training examples, with more weight on examples misclassified in previous rounds. This forces the base learner to focus on difficult cases. The weak rules are then combined, typically through a weighted majority vote. The chapter discusses the choice of base learning algorithm, but leaves it open to allow a general boosting procedure. Boosting is a provably effective method for combining rough rules into a highly accurate prediction rule. The chapter provides an overview of recent work on boosting, with a focus on AdaBoost, which has been extensively studied theoretically and empirically.