I am comparing 3 different classifiers ANN, XG Boost and Random Forest in making predictions. I also used SHAP for feature importance. I am only interested in the top 10 features based on SHA. The 3 classifiers share only 3 Top 10 features. ANN shares 5 features with Rf and 3 features with XG Boost and RF shares 7 features with XG Boost.
Additionally, I noticed that the range of the shap values differ significantly for the 3 models. for ANN, we have from -8 to 8. RF from -0.0.3 to 0.09 and XG boost from -.06 to 0.9.
I am not sure how to compare these models based on shap values. first i thought of focusing on the shared features between the 3 models before noticing the difference in shap values.
I guess my question is "What's the best way to compare these models based on their shap values? Does talking about the difference is shap values for the 3 classifiers make sense?
I have a picture for reference.
I found this Is it valid to compare SHAP values across models? but it doesn't seem to answer the questions I have.
