|
Abstract:
|
Given a fourth down scenario in a National Football League (NFL) game, we develop a method for evaluating whether the team on offense should attempt to gain a first down rather than punting or attempting a field goal. The approach is based on random forest methodology. One random forest, built on a training data set of third and fourth down plays from twelve NFL seasons, is used to obtain a distribution of outcomes for first down attempts. A second random forest is used to estimate the offense's win probability for each of the predicted outcomes. The expected win probability for the offense when attempting to gain the first down is then estimated by integrating win probability across the distribution of outcomes. This estimate of expected win probability is then compared to the offense's estimated win probability prior to attempting the first down to determine if going for the first down is a good choice. Several examples are presented to illustrate situations where going for the first down is (or is not) the recommended course of action.
|
