A game-predicting expert system using big data and machine learning

Abstract

The National Hockey League (NHL) is a major North American sports organization that earns $3.3 billion in annual revenue, and its stakeholders—team management, advertisers, sports analysts, fans, among others—have vested interest in league competitiveness and team performance. Utilizing player and team data collected from various web sources, we propose an expert system to better predict NHL game outcomes as well as improve recruiting and salary decisions. The system combines principal components analysis, nonparametric statistical analysis, a support vector machine (SVM), and an ensemble machine learning algorithm to predict whether a hockey team will win a game. The ensemble methods improve upon the reference SVM classifier, and the ensemble models’ predictive accuracy for the testing set exceeds 90%. The comparison of several ensemble machine learning approaches specifies opportunities to improve the accuracy of game outcome prediction. The system makes it simple for users to employ the learning methodologies and input data sources, evaluate model results, and address the challenges and concerns inherent in predicting hockey game wins.