Boosted Aggregate Models in Commerce and Industry: An Introduction and Case Study

Boosting is a means of constructing a strong classi er by aggregating a sequence of weaker classi ers. The sequence evolves by focusing the attention of a base learning algorithm on remaining examples that are hardest to classify. Boosting algorithms can thus be used to enhance the performance of a simple classi er (typically a regression tree or tree stump) in an automatic way. This makes boosting a particularly attractive approach in industrial settings. We use a simple simulated data set to illustrate that a boosted aggregate tree model can be preferable to a logistic regression or single tree model in some circumstances. We then perform cross-validatory tests on four real data sets typical of industrial and commercial problems, demonstrating that the generalisation performance of a regression tree model may often be improved using the Adaboost algorithm.