Interpretable Machine Learning with Python

Chapter 5: Global Model-Agnostic Interpretation Methods

In the previous chapter, Chapter 4, Fundamentals of Feature Importance and Impact, we demonstrated how permutation feature importance was a better alternative to leveraging intrinsic model parameters for ranking features by their impact on model outcomes. We also learned how to employ partial dependence plots and individual conditional expectation plots to examine how model outcomes change across feature values and interactions. However, even though all these global model-agnostic methods are exceedingly popular, they have something in common – they are sensitive to collinear features.

This chapter will continue looking at global model-agnostic methods, two of which were designed to mostly ...