Boosted Trees: A Powerful Toolbox for Statistical Analysis

Dr Glenn De'ath

Australian Institute of Marine Science,
Townsville, Queensland

Abstract

Boosted trees (BTs) are a statistical technique that can be used to explore, model and predict many types of data. BTs are widely recognised as one of the best modern data mining techniques yet they are flexible, powerful and easy to use. They challenge contemporary notions of statistical model selection, yet can be used to complement more traditional statistical methods such as generalised linear models (GLM) and generalised additive models (GAM).
This talk will outline the theory and application of BTs and how they relate to other treebased methods such as classification and regression trees, bagged trees and random forests. We will also discuss aggregated boosted trees (ABTs) - an enhanced version of BTs that provide additional functionality such as estimation of confidence intervals and better performance on small data sets. Examples of analyses using ABTs in R will be shown. In short, BTs are an extremely useful addition to the data analysis toolbox of all researchers.

Date: Friday 19th November 2010
Venue: Royal Society Room
Time: 12:00pm - 12:50pm