Machine learning has great potential for improving products, processes and research. But computers usually do not explain their predictions which is a barrier to the adoption of machine learning. This book is about making machine learning models and their decisions interpretable.

After exploring the concepts of interpretability, you will learn about simple, interpretable models such as decision trees, decision rules and linear regression. The 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 with Shapley values and LIME. In addition, the book presents methods specific to deep neural networks.

All interpretation methods are explained in depth and discussed critically. How do they work under the hood? What are their strengths and weaknesses? How can their outputs be interpreted? This book will enable you to select and correctly apply the interpretation method that is most suitable for your machine learning project. Reading the book is recommended for machine learning practitioners, data scientists, statisticians, and anyone else interested in making machine learning models interpretable.

Testimonials

"Thank you @ChristophMolnar for the great work on #MachineLearning Interpretability!" - @HJDLopes

"If you are looking for a good introduction to interpretable/explainable machine learning, this book is great. It covers lots of ground quickly and is well written, and is very up-to-date." - Tim Miller

"New book on interpretable #AI by @ChristophMolnar very much needed!" - @AjitJaokar

About The 2nd Edition

The 2nd edition of Interpretable Machine Learning offers a substantial improvement over the 1st edition. The book now also covers approaches specific to interpreting deep neural networks. The update also brings many new model-agnostic interpretation methods such as the popular SHAP, Anchors and functional decomposition. The 2nd edition also improves the arrangement of the chapters and fixes smaller typos and errors.

The Updates in Detail

All the methods chapters are now organized into four main chapters: interpretable models, global methods, local methods and deep learning specific methods. 

The following new chapters were added:

• Functional Decomposition
• Scoped Rules (Anchors)
• SHAP (SHapley Additive exPlanations)
• Preface by The Author

A new section about Neural Network Interpretation was added, containing the following (new) chapters:

• Learned Features
• Pixel Attribution (Saliency Maps)
• Detecting Concepts
• Adversarial Examples (already in 1st edition, but moved to this chapter)
• Influential Instances (already in 1st edition, but also moved to this chapter)

The following chapters got updated:

• Counterfactual explanations: The chapter was improved overall and Multi-objective counterfactual explanations were introduced.
• Partial Dependence Plots: A paragraph on PDP-based feature importance was added.
• Permutation Feature Importance:  The chapter was renamed, now lists alternative approaches, and a clear recommendation to use test data was added.
• Accumulated Local Effect Plots: Some additional disadvantages of the approach were added.
• Shapley Values: The explanation for conditional sampling  for Shapley values was improved.

Also, a few things were fixed:

• A mixup of the coding of the seasons in the bike rental data was fixed.
• Also the coding for cervical cancer outcome was fixed. This mostly impacts the logistic regression and the decision rules chapter.
• An error in the formula for R-squared was fixed.
• Many smaller typos and problems were. Here also a big thanks to all the readers who submitted fixes on Github!
• A lot of new software packages for interpretable machine learning are now available, and the 2nd edition now also lists newer candidates.

Other Versions

The print version can be bought here: http://bit.ly/iml-paperback

A free HTML version of the book can be found at: https://christophm.github.io/interpretable-ml-book/