I didn't know about SHAP graphs that's pretty cool. I'd love to be corrected but it just seems like an ensemble model of LightGBM, Gradient Boosting and Random

I didn't know about SHAP graphs that's pretty cool. I'd love to be corrected but it just seems like an ensemble model of LightGBM, Gradient Boosting and Random Forest with 12 features is just overkill. Why ensemble and not just 1 of the three models ? Much less likely to overfit. Looking at the average values BTS their speechiness values are so high compared to others I'm not sure whether those other features are adding anything, seeing featureImportance would be interesting (difficult to infer the relative importance of one feature to another since all we know is the ranking but not the values).