Contribute Media
A thank you to everyone who makes this possible: Read More

Why and how to explain machine learning predictions


Unfortunately, the predictive models that are most powerful are usually the least interpretable. However in some cases, for example fraud detection, end users need an understandable explanation of a particular prediction, at observation level, and not only at population level (e.g. : features importance). During this talk we will present different approaches to tackle this issue, both for random forests and gradient boosting trees. We will also demonstrate an implementation based on scikit-learn.


Improve this page