Introduction Method Goal Results References Conclusion Abstract
Undergraduate Category: Engineering and Technology Degree Level: Undergraduate Abstract ID#: 1026
Machine Learning on Sports Prediction Samuel Starkman, Emerson Boyd, Jeremy Eng, Kevin Kimelman, Xiangyu Li, and David Kaeli
Abstract
Method
● Sports data analysis frameworks are created for many purposes, including: ○ Informing coaches for better decision making ○ Improving match prediction accuracy ○ Maximizing stadium revenue
Results ● The prediction for the 2012 season generated using a Gaussian Process model: ○ 56.99% of the time, the model selects the correct game winner ○ 81.82% of the time, the model succeeds in placing a winning bet
● Offensive and defensive statistics from the 2003 to 2012 NFL season were used to build an accurate prediction model
Introduction ● Current data frameworks are incomplete or inaccurate ● A Gaussian Process is used to create a model from the data ○ This model is used to predict the margin of victory for each game ○ Using the predicted margin of victory and the Las Vegas line, an intelligent bet can be made
Goal
● Acquire statistics for each team and the Vegas betting line for each game from 2003-2012 ● Apply the Gaussian Process on every game for seasons 2003-2009 to create a proper model ○ We will refer to this data as the “training set” ○ We evaluate four key game statistics ● Using this model, and validating the model with data from 2010-2011, we test the validated model against games from 2012
Margin of Victory Error
Game Winner Accuracy
Vegas
10.69
64.71%
Gaussian
11.65
56.99%
Bets Made
Bets Won
Winning Ratio
11
9
81.82%
Conclusion ● While the model is still in a developmental stage, it already provides very respectable accuracy ● In the future, information on other significant factors (e.g., injuries) can be included in the training set to increase the accuracy of our model
● Predict the margin of victory in NFL games ● Develop betting schemes that can maximize the chances of winning ○ “Beat” the Las Vegas Line ● Utilize this framework for general purpose applications outside of sports
References [1] Jim Warner. Prediction Margin of Victory in NFL Games: Machine Learning vs. the Las Vegas Line, Dec 17, 2010. [2] C. E. Rasmussen and C. K. I. Williams. Gaussian Processes for Machine Learning. The MIT Press, Cambridge, MA, 2006. [3] Vegas Lines: http://www.repole.com/sun4cast/data.html
Machine Learning on Sports Prediction Samuel Starkman, Emerson Boyd, Jeremy Eng, Kevin Kimelman, Xiangyu Li, and David Kaeli
Abstract
Method
● Sports data analysis frameworks are created for many purposes, including: ○ Informing coaches for better decision making ○ Improving match prediction accuracy ○ Maximizing stadium revenue
Results ● The prediction for the 2012 season generated using a Gaussian Process model: ○ 56.99% of the time, the model selects the correct game winner ○ 81.82% of the time, the model succeeds in placing a winning bet
● Offensive and defensive statistics from the 2003 to 2012 NFL season were used to build an accurate prediction model
Introduction ● Current data frameworks are incomplete or inaccurate ● A Gaussian Process is used to create a model from the data ○ This model is used to predict the margin of victory for each game ○ Using the predicted margin of victory and the Las Vegas line, an intelligent bet can be made
Goal
● Acquire statistics for each team and the Vegas betting line for each game from 2003-2012 ● Apply the Gaussian Process on every game for seasons 2003-2009 to create a proper model ○ We will refer to this data as the “training set” ○ We evaluate four key game statistics ● Using this model, and validating the model with data from 2010-2011, we test the validated model against games from 2012
Margin of Victory Error
Game Winner Accuracy
Vegas
10.69
64.71%
Gaussian
11.65
56.99%
Bets Made
Bets Won
Winning Ratio
11
9
81.82%
Conclusion ● While the model is still in a developmental stage, it already provides very respectable accuracy ● In the future, information on other significant factors (e.g., injuries) can be included in the training set to increase the accuracy of our model
● Predict the margin of victory in NFL games ● Develop betting schemes that can maximize the chances of winning ○ “Beat” the Las Vegas Line ● Utilize this framework for general purpose applications outside of sports
References [1] Jim Warner. Prediction Margin of Victory in NFL Games: Machine Learning vs. the Las Vegas Line, Dec 17, 2010. [2] C. E. Rasmussen and C. K. I. Williams. Gaussian Processes for Machine Learning. The MIT Press, Cambridge, MA, 2006. [3] Vegas Lines: http://www.repole.com/sun4cast/data.html
