Data Mining for Business Analytics

CHAPTER 13 Combining Methods: Ensembles and Uplift Modeling

In this chapter, we look at two useful approaches that combine methods for improving predictive power: ensembles and uplift modeling. An ensemble combines multiple supervised models into a “super-model.” The previous chapters in this part of the book introduced different supervised methods for prediction and classification. Earlier, in Chapter 5, we learned about evaluating predictive performance, which can be used to compare several models and choose the best one. An ensemble is based on the powerful notion of combining models. Instead of choosing a single predictive model, we can combine several models to achieve improved predictive accuracy. In this chapter, we explain the underlying logic of why ensembles can improve predictive accuracy and introduce popular approaches for combining models, including simple averaging, bagging, and boosting.

In uplift modeling, we combine supervised modeling with A–B testing, which is a simple type of a randomized experiment. We describe the basics of A–B testing and how it is used along with predictive models in persuasion messaging not to predict outcomes but to predict who should receive which message or treatment.

Python

In this chapter, we will use pandas for data handling and scikit-learn for the models. We will also make use of the utility functions from the Python Utilities Functions Appendix. Use the following import statements for the Python code in this chapter.

 import ...