This is a review of the research conducted by Xinyu Wei, Patrick Lucey, Stuart Morgan, Machar Reid, and Sridha Sridharan applying random decision forest and other sports analytics methods.
Strategy is the name of the game in tennis with players continually attempting to maneuver their opponents into weak positions in order to score points. The final shot is not the most important element of the rally, but rather the strategy preceding the final shot. This research looks for the strategy behind the progression of a tennis rally used by top players.
Players systematically hit the tennis ball in a manner that will put their opponent in a weaker position, thus gaining dominance for themselves. Players that use similar strategies can be grouped together according to style. It is also important to look at the series of shots in a rally to determine context.
A player will work towards moving their opponent to a specific position on the court, and once this is achieved, they gain an advantage and are able to win the point.
Three years of data from matches in the Australian Open Men’s singles draw, including 2292 winners and 37727 shots are analyzed, focusing on the top 10 players who played the most matches. Information regarding the position of the player, position of the ball, current score, point duration, server, and receiver identity are included.
Raw features such as the trajectory of the ball are included as well as dominance features including ground stroke speed ratio, ground stroke weight ratio, and lateral player movement ratio.
A Random Decision Forest is used to predict the probability that the next stroke is likely to be the winner. This provides a universal model for all tennis players. However, individual players each have their own unique style. A tennis player’s style is discerned through learning a dictionary of shot trajectories from data by optimizing prediction performance and reconstruction error. This dictionary includes single shot elements as well as shot combinations. Shot combinations are grouped using a K-means algorithm. Evaluating style is then conducted via a reconstruction error and prediction loss process based on both single shot and shot combination information.
Context information is used to describe elements of the rally occurring before the point is scored. This includes the score line, elapsed match time, environment conditions such as wind and temperature, and the court surface.
Style and context combined together provides superior information for predicting the winner of the rally.
Analysts can this tool to make predictions during a rally as to the probable outcome. They can also compare and contrast the strategies used by the various players. Even when two players have rarely met before, or never met, analysts can look at players with similar styles to predict the outcome of the game. Coaches and players can use the style groupings to analyze future opponents, allowing them to prepare the best possible strategy.
Incorporating style, context, single shot, and shot combinations provides an increasingly accurate ability to predict the outcome of a shot in a tennis game.
Analytics Used: Root mean square error, ground stroke weight ratio, lateral player movement ratio, dictionary learning, random decision forest
Find out how Sports Analytics Expert Victor Holman can give your team the competitive advantage.
How mature is your team’s analytics program? Take the Sports Analytics Maturity Assessment.
Discover the Groundbreaking Sports Analytics Application and Framework coaches and sports analysts are talking about!
Learn all about sports analytics in Victor Holman’s Sports Analytics Blog.