Interpretability in Machine Learning
Marco Repetto · 2023
From self-driving cars to score credit ratings, Machine Learning (ML) is already starting to shape our world. This chapter aims to provide a broad overview of interpretable ML. It covers why interpretability is essential, the different approaches to interpreting ML models, and the challenges involved in making ML models interpretable. The chapter provides the reader with some knowledge of the recent implementations of such techniques, either in Python or R. The history of interpretability in ML goes back to work in the early days of AI and cybernetics. In the early days, the field was primarily concerned with methods for analyzing and understanding the behavior of linear models. Interpretability is essential in ML for several reasons and can benefit many different stakeholders. The goal of interpretability is to provide a model that can be given to a decision-maker to understand how it is making its predictions. Essentially, interpretability provides a human-friendly description of how the model makes its decisions.