Author_Institution :
Dept. of Bioeng., Pittsburgh Univ., PA, USA
Abstract :
In the National Football League (NFL), teams that outperform their opponents in four categories usually are victorious. These statistics are yards gained, rushing yards gained, turnover margin, and time of possession. In fact, through the first eight weeks of the 1994 regular season, no team leading its opponent in all four categories had lost. In general, the more categories a team leads, the greater its chance of winning a game. Therefore, the relative strength of NFL teams can be established by comparing these four statistical categories. Several neural network strategies are tried to predict winners in NFL games. Binary, ternary, and continuous input vectors are used as inputs to appropriate networks: Hamming, adaptive resonance theory (ART), Kohonen self-organizing map (SOM), and backpropagation (BP). Predictions are presented, and the performance of each network is examined. Network results using supervised and unsupervised training methods are also compared
Keywords :
ART neural nets; backpropagation; self-organising feature maps; sport; statistical analysis; Hamming networks; Kohonen self organizing map; NFL teams; National Football League; adaptive resonance theory; backpropagation; binary input vectors; continuous input vectors; game; neural network performance; neural network quarterbacking; neural network strategies; possession time; rushing yards gained; statistics; supervised training methods; ternary input vectors; turnover margin; unsupervised training methods; winners prediction; yards gained; Adaptive systems; Multi-layer neural network; Multilayer perceptrons; Neural networks; Neurons; Organizing; Statistics; Subspace constraints;
