Slide 1ObjectivePurposeDataWhat-if AnalysisPreliminary What-if results:Predicting the 2010-2011 SeasonQuestions?Predicting NFL Team Performance, a Closer Look at Fantasy FootballECE 539Presented: 12/14/2010Joseph QuigleyObjectiveTrain a multi-layer perceptron network to predict the regular season records of NFL Football teams. (Within a range.)Wins in a season:0-34-89-1213-16PurposeCreates a simple way to turn projected fantasy football statistics into projected wins and losses.To have the ability to create hypothetical teams and estimate how many games they would win in a season.What if the 2008 Lions (winless) went back in time and traded defenses with the 2000 Ravens (one of the best defenses in recent history)?What if the 2007 Patriots (only 16-0 team ever) traded defenses with the 2006 Redskins (one of the worst defenses in recent memory, and in the last 10 years of fantasy football)?DataFantasy Football statistics (2005-2010):QuarterbackRunning backWide ReceiverTight EndKickerDefense/Special TeamsTeam VectorWhat-if AnalysisCan’t just add another teams fantasy defense/ST value.Needed to modify offensive production based on turnovers.Calculated number of offensive possessions in a season, then the fraction of fantasy points per possession.Multiplied this by the difference in turnovers between the new and the old defense.Preliminary What-if results:2008 Lions/2000 Ravens – Win: 6.6-9.6 games2008 Lions/2000 Broncos – Win: 7.4-10.4 games2007 Patriots/2006 Redskins – Win: 6.4-9.6 games.Predicting the 2010-2011 Season0-3 4-8 9-12 13-160246810121416182007-2009 AverageAverage PredictionNumber of WinsNumber or
View Full Document