160622.pdf

Abstract

With new shot-tracking technology and improved data-gathering tools, tennis has become a field where sports analytics can be utilized for match predictions. At the same time, tennis has become the second largest market in a sports betting industry worth an estimated $833 million as of 2019. By developing accurate models through data analytics, there exists an opportunity to take advantage of inefficiencies in the market and profit off of the large volume of bets placed on matches. This paper will explore various models and features that can best predict win probabilities of tennis, including features that can be gathered from a novel tennis dataset with point-by-point text descriptions. This thesis evaluates the accuracy of models such as logistic regression, random forest regression, and support vector machines in predicting in-game updating win statistics. We find that using a logit model trained on match scoring and in-depth player statistics outperforms previous models in generating tennis win probabilities.