Predicting Tennis Match Outcomes with Network Analysis and Machine Learning

Singles tennis is one of the most popular individual sports in the world. Many researchers have embarked on a wide range of approaches to model a tennis match, using probabilistic modeling, or applying machine learning models to predict the outcome of matches. In this paper, we propose a novel approach based on network analysis to infer a surface-specific and time-varying score for professional tennis players and use it in addition to players’ statistics of previous matches to represent tennis match data. Using the resulting features, we apply advanced machine learning paradigms such as Multi-Output Regression and Learning Using Privileged Information, and compare the results with standard machine learning approaches. The models are trained and tested on more than 83,000 men’s singles tennis matches between the years 1991 and 2020. Evaluating the results shows the proposed methods provide more accurate predictions of tennis match outcome than classical approaches and outperform the existing methods in the literature and the current state-of-the-art models in tennis.