Longitudinal Comparison of Models to Predict Match Outcomes in the WTA

Abstract

Analytics have been less utilized in women’s professional tennis (WTA), compared to other professional sports. Despite unique difficulties in predicting match outcomes, there has been a spate of recent articles that utilize prediction tools applied to men’s profession tennis (ATP) data. Our research adds efficiencies and new features to previously-created probabilistic models for longitudinal predictions of WTA matches. We compute, update, and analyze a set of related summary statistics along with specific match details for individual players and integrate these with Bradley-Terry algorithmic modeling of match probabilities to incorporate strength of schedule. Data for player statistics and results of WTA tournaments was obtained from a GitHub repository under a Creative Commons license. We edited and created original functions in R: wrangling the data across an appropriate time window, court surface, and player rank; and implementing an existing algorithm for prediction and assessment. We also apply Elo ratings for comparative prediction, utilizing a longitudinal update and weighting by strength of win. We discuss the methods and coding, and apply elevated error analysis of match predictions compared to observed match outcomes to determine the overall accuracy of our model; accurate predictions could further inform the ranking of WTA players.