Support Vector Machines

predict() and predict_proba() >= 0.5 disagreed on 25 of 3,000 training rows, and the model kept 48.2% of them as support vectors.