NBA Game Predictions based on a Team Composition Model

Abstract

Over the years the sports industry has grown into a multi-billion dollar industry in which technology plays an essential role. As the industry started collecting more data, the research into analyzing that data advanced simultaneously. This thesis will focus on developing a model that is going to be used by several learning algorithms to predict the game outcome of future games. The model will extend upon the results of Singh [16] to measure the effect of adding personalized statistics for each player and use team-compositions to predict the outcome of each game. Research shows that Machine learning algorithms perform better than traditional statistical models when trying to capture a latent context or in this case team-composition. Therefore, the learning is done by a set of supervised learning algorithms which consist of Logistic Regression, Neural Networks and Linear Support Vector Machines.