Interpretable AI

3 Model-agnostic methods: Global interpretability

This chapter covers

• Characteristics of model-agnostic methods and global interpretability
• How to implement tree ensembles, specifically random forest—a black-box model
• How to interpret random forest models
• How to interpret black-box models using a model-agnostic method called partial dependence plots (PDPs)
• How to uncover bias by looking at feature interactions

In the previous chapter, we saw two different types of machine learning models—white box and black box—and focused most of our attention on how to interpret white-box models. Black-box models have a high predictive power and, as the name suggests, are hard to interpret. In this chapter, we will focus on interpreting black-box models, ...