Consider the correlation between multiple features (look up correlation plots and sns. heatmap). Try to show redundancy in a group of features (e. g. body and html might co-occur relatively frequently, or you might be able to design a feature that captures all html tags and compare it to these). Visualize which words have high or low values for some useful statistic. Visually depict whether spam emails tend to be wordier (in some sense) than ham emails