Cracking the Box: Interpreting Black-Box Machine Learning Models

Practical techniques for keeping production ML models interpretable and trustworthy.

As AI and machine learning are adopted across nearly every industry, keeping models interpretable becomes critical. The decisions algorithms make should be understandable to people and to the business — both to avoid bias and to manage uncertainty.

Because businesses have to be able to explain the decisions their ML models take, black-box models present a unique challenge for the teams that build and run them.

What’s inside

Classes of interpretability methods
Methods of ML interpretability, including partial dependency plot, permutation importance, SHAP, LIME, and Anchor
Pros, cons, features, and specifics of each ML interpretability method

Data science and ML engineering leaders responsible for putting models into production in regulated or high-stakes domains — and the business stakeholders who need to trust and defend those decisions.