Found Something

Posted by Shirl Williams on May 14, 2020

Models are funny things. You can use all manner of calculus, statistics, and even linear algebra to shape and reshape data into something unrecognizable but still have a way to make predictions with certain accuracies. In an effort to make sense of the model I ran across something even my instructor had not seen. SHAP (SHapley Additive exPlanations) is a game theoretic approach to explain the output of any machine learning model. It can also break down a prediction to show the impact of each feature. This leads to some amazing visualizations. After this boot camp, I will need to explore modeling with SHAP.

More information and visualization examples in [articles] like this (https://towardsdatascience.com/explain-any-models-with-the-shap-values-use-the-kernelexplainer-79de9464897a). I’ve heard there’s even a YouTube video for those interested in learning SHAP.