Chapter 3: Feature Effects
Feature Effects indicate the change in prediction due to changes in feature values. This chapter explains the feature effects methods ICE curves, PDP and ALE plots.
-
Chapter 3.1: Introduction to Feature Effects
We have already learned that interpretation methods can be divided into local and global ones. In this section, we embed feature effects into this framework.
-
Chapter 3.2: Individual Conditional Expectation (ICE) Plots
Individual conditional expectation (ICE) curves are a local feature effect method. The idea is to replace the values of the feature of interest while keeping all other features unchanged and see how the prediction varies for each observation.
-
Chapter 3.3: Partial Dependence (PD) Plot
Averaging ICE curves leads to a global interpretation method called partial dependency plot (PDP). Besides the relationship between PDP and ICE curves, the centered version of ICE plots (c-ICE) is also discussed.
-
Chapter 3.4: Accumulated Local Effect (ALE) Plot
PDPs suffer from problems with extrapolation and correlation. One workaround is marginal plots (M-plots), though these in turn suffer from omitted variable bias. ALE (Accumulated Local Effects) diagrams cope with all these complications.