Overview

Sharpen your decision tree expertise during this one-day advanced course, geared towards analysts and modelers with prior knowledge of tree algorithms. Using case studies, seminar topics include:

• Hybridizing CART with logistic regression.
• Combining multiple trees via bagging, boosting, repeated sampling and varying priors.
• Understanding the strategy behind alternative splitting rules and their characteristic strengths, weaknesses and signatures.
• Learning how CART implements variable costs of misclassification; separating growing with costs from pruning with costs.
• Emerging topics in decision trees; highlights of recent developments.

Content and instructional methods

Attendees will see examples of analysis of real world data. PowerPoint slides and live modeling runs will facilitate the learning process.

Course Outline:

1. Binary recursive partitioning
2. Splitting rules
   1. Gini
   2. Entropy
   3. Twoing
3. Fundamental tree building mechanisms
   1. Priors
   2. Costs
4. Model evaluation using relative costs and ROC
5. Hybrid modeling
   1. CART and LOGIT
   2. CART and Linear Regression
6. Large trees and their properties
7. Ensemble models
   1. Bagging
   2. Arcing
8. Novel uses of CART