Trained on the dataset ENRON-spam, train/test split 80/20, seed 42