Math and Data

This book review is for Interpretable Machine Learning: A Guide for Making Black Box Models Explainable, by Christoph Molnar, the second edition of which was self-published in 2022 and spans 319 pages. The book is about making machine learning models and their decisions interpretable, which is a significant issue because computers do not usually explain their predictions, hindering the adoption of machine learning.

The lack of interpretability of machine learning models has been a major obstacle in their adoption, despite their great potential for improving products, processes, and research. Computers are not transparent about how they arrive at their predictions, and this can create a lack of trust in their decisions. This is especially problematic in high-stakes applications such as medical diagnosis, where a wrong prediction could have serious consequences.

Fortunately, there has been a growing interest in making machine learning models interpretable, and this is the focus of the book Interpretable Machine Learning by Christoph Molnar. The author recognizes that many existing models are considered “black boxes” because their internal workings are not transparent, and he sets out to provide practical tools for interpreting them.

The book begins with exploring the concept of interpretability, followed by a discussion of simple, interpretable models such as decision trees, decision rules, and linear regression. The main focus of the book is on model-agnostic methods for interpreting black box models, such as feature importance and accumulated local effects, and explaining individual predictions using Shapley values and LIME. In addition, the book presents methods specific to deep neural networks.

Each interpretation method is explained in depth, with a critical evaluation of their strengths and weaknesses, and how their outputs can be interpreted. The book enables machine learning practitioners, data scientists, statisticians, and anyone interested in making machine learning models interpretable to select and apply the most appropriate interpretation method for their project.

The book appears as a collection or little encyclopedia of various methods and model performance metrics, with a focus on exhaustivity over selectivity. The level and amount of mathematics are appropriate, offering a real feeling of what the techniques do.

The author provides applications based on real or “prototype” data, with source code on the author’s GitHub repository and website. You can also purchase the published version on Amazon. However, looking through the reviews, the print version’s quality seems to have room for improvement, and a PDF version with clickable links would be preferable.

The book is highly recommended for machine learning practitioners, data scientists, statisticians, and anyone interested in making machine learning models interpretable. The book provides a comprehensive and critical evaluation of interpretation methods for machine learning models, making it a valuable resource for those seeking to improve their understanding of interpretability in machine learning and its practical applications.