I used the enron email corpus to create features that were then used to predict POI (persons of interest) who might have been involved in fraud.
Since this project I have learned many new machine learning techniques that I know would perform better than what I have here. Namely, I believe SVM and/or random forest would be a better classifer. I used Naive Bayes here as it performed surprisingly well.
I'd also like to work on feature engineering. It's possible there are features I could find that might work better than what I used in my model, although my pipeline does include select k-best and a gridsearch that brought the performance of the model to a level I was happy with.