Research Mentor(s)

Brian Hutchinson

Affiliated Department

Computer Sciences

Sort Order

52

Start Date

14-5-2015 10:00 AM

End Date

14-5-2015 2:00 PM

Keywords

Baseball, Expectation maximization, Recurrent neural networks, Modeling

Document Type

Event

Abstract

The 162 game long Major League Baseball season provides ample time for a player’s performance to vary and trend in different directions. Managers must set daily rosters for their teams, using past performance to help make decisions. But which prior performance periods tell us the most about upcoming performance? To answer this, it's helpful to view a player’s future performance, for any given statistic, as a function of his performance in previous playing periods (e.g. previous game, previous week, previous year, etc.). In this on-going research project, we consider two approaches to predicting future performance from the past. In the first, we build a probability mass function for each of a set of discrete, disjoint past time periods and we use Expectation Maximization to learn the appropriate weights for each period to best predict future outcomes. In our second approach, we predict a player's performance in the next game based on all previous history using a recurrent neural network.

Share

COinS
 
May 14th, 10:00 AM May 14th, 2:00 PM

Time Series Modeling of Baseball Performance​

Computer Sciences

The 162 game long Major League Baseball season provides ample time for a player’s performance to vary and trend in different directions. Managers must set daily rosters for their teams, using past performance to help make decisions. But which prior performance periods tell us the most about upcoming performance? To answer this, it's helpful to view a player’s future performance, for any given statistic, as a function of his performance in previous playing periods (e.g. previous game, previous week, previous year, etc.). In this on-going research project, we consider two approaches to predicting future performance from the past. In the first, we build a probability mass function for each of a set of discrete, disjoint past time periods and we use Expectation Maximization to learn the appropriate weights for each period to best predict future outcomes. In our second approach, we predict a player's performance in the next game based on all previous history using a recurrent neural network.

 

To view the content in your browser, please download Adobe Reader or, alternately,
you may Download the file to your hard drive.

NOTE: The latest versions of Adobe Reader do not support viewing PDF files within Firefox on Mac OS and if you are using a modern (Intel) Mac, there is no official plugin for viewing PDF files within the browser window.