What is the SHAP beeswarm chart? How can I best interpret this chart?

Answer:

The SHAP beeswarm chart is a feature impact visualization and can be found in the “Feature Analysis” section of the One AI Results Summary report (One AI >Runs > Run Label > Results Summary).  SHAP, as explained in the previous question, provides numeric importance values for every feature for every individual prediction.  The beeswarm plots every one of these values as a dot on a chart.  The horizontal axis indicates how predictive that feature is for that individual in a positive (= positive label) or negative direction.  In addition, the beeswarm chart adds color coding of each dot based on how the feature value for that individual compares to the average for the entire population. 

For categorical features such as “is_future_manager” in the image below, that is easy to see.  1.0 is red and 0.0 is blue.  The “is_future_manager” of this chart paints a clear picture that someone being a future manager is a powerful indicator of them terminating.  For scaled features, the coloring is a gradient.  For date of birth, 2000-01-01 might be red, 1985-01-01 purple, and 1970-01-01 blue.  Remember that SHAP must be enabled in the global settings for the model prior to running in order for the beeswarm to be available for that run.

 

SHAP Beeswarm Chart in the Results Summary Report

Was this article helpful?

2 out of 2 found this helpful

Comments

0 comments

Please sign in to leave a comment.