Exploration: Feature Weighting in a Prediction Models

Say we're creating a machine learning model to predict successful orators in a college Speech class. We might take video recordings of successful orators in past classes and code each for various features. When millions of features are used in prediction, it's useful to be able to gain insights into those which have the highest impact.
These visualizations explore ways of highlighting the correlation of an orator saying 'differentiate' in their speech and that orator being a successful orator.
Hint: the bottom right is the most informative of these four.