130 likes | 202 Views
Prediction of NBA games based on Machine Learning Methods. University Of Wisconsin Madison. ECE/CS/ME 539 Introduction to Artificial Neural Networks and Fuzzy System Renato Amorim Torres December, 2013. Proposal. The objectives of the project are: Predict the winning team of a NBA game
E N D
Prediction of NBA games based on Machine Learning Methods University Of Wisconsin Madison ECE/CS/ME 539 Introduction to Artificial Neural Networks and Fuzzy System Renato Amorim Torres December, 2013
Proposal The objectives of the project are: • Predict the winning team of a NBA game • The goal is to have the Prediction Rate higher than the rate of the Very Naive Majority Vote Classifier. This method looks at all previous games (in the season) of the two teams and picks the team with the fewest losses as the winner.
Preparation of the Data • Data was obtained on the website: www.basketball-reference.com
Preparation of the Data • Box scores were copied into a spreadsheet • Using macro all the team names were replaced by numbers, the unnecessary columns were deleted, and .txt files were generated with the data. • The .txt files were loaded in MatLab and used to generate the feature vectors and implement the methods.
Data Analysis Blue Line >> 2012 to 2006 Regular Seasons Red Line >> Mean of Seasons
Feature Vectors • Win-Loss percentage of both teams • Point differential per game of both teams • Win-Loss percentage in the previous 8 games for both teams • Visitor Team win-Loss percentage as visitor • Home Team win-Loss percentage at home Total of eight features
Feature Vectors • Win-Loss percentage in the previous 8 games of both teams. Why eight games? Blue Line >> 2012 to 2006 Regular Seasons Red Line >> Mean of Seasons
Prediction • Based on the graph shown before, it was found that each team should have played at least eight games prior to predicting games. • Considering this starting point, the Naive Majority Vote Rate was calculated in order to define the goal:
Prediction Maximum Likelihood Classifier - Not all the feature vectors were used. In order to establish the best features, a code was implemented to verify the prediction rate for all combinations of features. Features used: • Feature #2: Home Team Win-Loss percentage at home • Feature #4:Home Team total Win-Loss percentage • Feature #5:Visitor Team point differential per game in the actual season • Feature #7/8: Win-Loss percentage in the last 8 games for both teams
PredictionMaximum Likelihood Classifier -The training data was a combination of the previous season's games and the testing data was random games of the “current” season. The likelihood classifier achieved the following results:
PredictionMaximum Likelihood Classifier Results For comparison... Table extracted from: Matthew Beckler, Hongfei Wang. and Michael Papamichael NBA Oracle
Prediction Linear Regression / LMS Algorithm - In order to find the weights, the LMS Algorithm was implemented: