Our final accuracy was 66.9922%. This was just .0078 below our initial goal of getting within 5% of the current best NFL predictors. We benefited from DTNB as a classifier (initial accuracy ~65%) and then added another ~2% accuracy by defining and coding our own mathematical expressions.
Another interesting thing we learned from our testing was that yard differentials have essentially no effect on game outcome. Turnovers, for example, were far more important. In other words, it doesn’t matter how far a team drives downfield if they fail to convert in the Red Zone or turn the ball over.