Interpretable machine learning products

Description

Interpretable models result in more accurate, safer, and more profitable machine learning products. A model you can interpret and understand is one you can more easily improve and can offer insights that can be used to change real-world outcomes for the better. It is also one you, regulators, and society can better trust to be safe and nondiscriminatory. But interpretability can be hard to ensure. There is a central tension between accuracy and interpretability: the most accurate models are necessarily the hardest to understand.

Michael Lee Williams explores the growing business case for interpretability and its concrete applications, including churn, finance, and healthcare. Along the way, Michael offers an overview of the open source, model-agnostic tool LIME, which gets around the accuracy-interpretability tension by allowing you to peer inside black-box models. Michael concludes with a demonstration of a working web application that uses LIME to explain why customers churn and raises the possibility of intervening to prevent their loss.