Msis5633 cfb presentation

Predictive Measures in College Football

By David AffentrangerMSIS 5633 OSU Fall 2012

Project Outline

Project Research

Data Sources and Tools

Data Transformation

Decision Tree

Decision Tree Results

Classification Predictive Modeling

Prediction Results

Future Efforts

Project Research

Projecting Point Spreads◦ http://cs229.stanford.edu/proj2010/LiuLai-BeatingTheNCAAFootballPointSpread.pdf

Predicting individual games◦ http://cfbpredictions.com/

Predicting BCS Rankings◦ http://harvardsportsanalysis.wordpress.com/2011/

11/24/making-sense-of-the-chaos-a-bcs-prediction-model/

Data Sources and Tools

Data Sources◦ http://www.cfbstats.com

Tools◦ Microsoft Excel

◦ Rapid Miner

Data Transformation

Data Challenges◦ Data home/away teams listed in separate spreadsheet

◦ Data rows lacked visiting team statistics

◦ Data rows lacked who won the game

◦ Many numerical fields difficult for classification

Data Solutions◦ Used Excel vlookups to pull in visiting team statistics into

the tuples

◦ Used the “Points” field to compare home vs away to create a “Winner” classification field.

◦ Built Formula fields to transform numerical values into a text classification (TOP, Special Teams, Penalty Yards)

Data Transformation

Decision Tree

Using Rapid Miner and a Decision Tree (Gini Index) built a model to determine key factors for winning

Decision Tree Results

Visit Rush Yards Most important Factor in the model >= 165 Yards Visit Winner 66% of the time

<= 165 Yards Home Winner 79% of the time

Decision Tree Results

Classification Predictive Modeling

Predictive Model Plan◦ Predict winner of an individual game based upon key statistics

◦ Use same data set as the decision tree

◦ Split the data into Testing and Training

◦ Evaluate different data elements and there affects on prediction

Classification Predictive Modeling

Using Rapid Miner◦ Naïve-Bayes Classification Model

◦ Split Example set into 2 sets using X-Validation

◦ Analyzed using Performance Tool

Prediction Results

Prediction Results

85.58% Home Winner precision

78.93% Visit Winner precision

Future Efforts

Model currently relies on a user manually entering key statistical information for a game

Model could be enhanced for a full predictive solution ◦ Using a neural network statistical data could be predicted for an individual team for a particular game

◦ Data from the neural network could feed the classification model to predict a game or series of games

References

http://www.cfbstats.com

http://espn.go.com/college-football/

http://cs229.stanford.edu/proj2010/LiuLai-BeatingTheNCAAFootballPointSpread.pdf