Paul McBride
ECE 539
Final Project Proposal
Football is the most popular sport in America. It attracts millions of people across the nation to tune into the action every sunday. And just like every other professional sport, it is becoming more and more about statistical analysis. Front office personnel, head coaches, and assistant coaches are constantly analyzing game play stats to see what they can add to the team to achieve more victories. Betting agencies are also analyzing stats trying to come up with the correct odds to place on a particular game. Millions of dollars are riding on these games every week. I want to delve into the stats to find out the best predictor to winning an NFL game.
I will try to predict the winner of an NFL game by applying NFL game stats to a neural network. I will use a back-propagation multilayer perceptron. A multilayer perceptron can handle a large number of cases, both linear and nonlinear and back-propagation will allow me to use some form of supervised learning. This neural network will be useful for this type of data set, which is essentially a classification problem; win or lose.
The stats that I will use are the combined team stats such as total offense, turnovers, and quarterback rating. This day in age, there are many, many sets of team play data on a week by week basis. I plan on getting my data from NFL.com. They have reliable and accurate information in tables that I can pull from. If that proves too difficult I will search for NFL stats that come prepared in a CSV (comma separated value) format. That will allow me to easily import the data into an excel sheet. I will use primarily offensive stats when preparing my feature vectors of which there are very many. To boil down the feature, I will construct the ROC curves of each attribute to pick out roughly 4 or 5 important attributes to use for my feature vectors.
References:
http://www.repole.com/sun4cast/data.html For CSV files of data.
http://www.nfl.com To check if the data is reliable.
Purucker, Michael C; Neural network quarterbacking: Who different training methods perform in calling the games. IEEE Potentials, August/September 1996. http://ieeexplore.ieee.org/iel1/45/11252/00535226.pdf?arnumber=535226