Chapter 4: Fundamentals of Feature Importance and Impact
In the first part of this book, we introduced the concepts, challenges, and purpose of machine learning interpretation. This chapter kicks off the second part, which dives into a vast array of methods that are used to diagnose models and understand their underlying data. One of the biggest questions answered by interpretation methods is: What matters most to the model and how does it matter? Precisely, interpretation methods can shed light on the overall importance of features and how they—individually or combined—impact a model's outcome. This chapter will provide a theoretical and practical foundation to approach these questions.
In this chapter, we will first use several scikit-learn models' intrinsic parameters to derive the most important features. Then, realizing how inconsistent these results are, we will learn how to use Permutation Feature Importance (PFI) to rank the features intuitively and...
