ABSTRACT
This chapter is dedicated to select topics from the increasingly popular field of interpretable machine learning, which easily deserves its own book-length treatment, and it has; see, for example, Molnar and Biecek and Burzykowski. Tree-based ensembles, especially the ones discussed in the next two chapters, can provide state-of-the-art performance, and are quite competitive with other popular supervised learning algorithms, especially on tabular data sets. The permutation approach uses the difference between some baseline performance measure and the same performance measure obtained after permuting the values of a particular feature in the training data. Although permutation importance is most naturally computed on the training data, it may also be useful to do the shuffling and measure performance on new data.
