2016 • Engineering and Technology
Machine Learning on Sports Prediction
Lead Presenter: Samuel Starkman
PI: David Kaeli
Sports are an integral part of today’s mainstream media and society, appearing virtually everywhere. Whether sports are considered jobs, ways of life, or simply pure entertainment, they are a universal language. Today, sports are no longer just being watched and played. With recent technological advancements, such as wearable trackers and motion cameras, sports data is now being rapidly generated and shared. As a result, many data analysis frameworks have been developed in order to assist in all facets of sports, including helping coaches make decisions, improving match prediction accuracy, and maximizing stadium revenue.
In this project, machine learning algorithms are applied to predict the outcomes and margins of victory in National Football League (NFL) games. Machine learning techniques are applied on large amounts of data obtained from various official NFL websites. This training data is comprised of game-by-game data from the 2003 to the 2014 season, containing a variety of offensive and defensive statistics. The most influential features are then selected to build an accurate prediction model using Gaussian Processes.