Elite sport is now awash with data. As athletes and management look to gain every competitive advantage they possibly can, they are gathering information about all aspects of individual and team performances in both training and matchplay, as well as a raft of other metrics.
One tool analysts are using to transform the data being gathered into actionable insights is machine learning.
Machine learning is a form of artificial intelligence which is able to learn without a human programming it to specifically find something. Machine Learning algorithms do this by searching large data sets for meaningful patterns, from which future events can be predicted or classified. It finds the sort of patterns that are often imperceptible to traditional statistical techniques because of their apparently random nature.
The confidence often needed to get coaches to the pinnacle of their field means that some are still reluctant to cede ground to algorithms and machines, but inherent prejudices and the fallibility of human memory mean that the brain is an inefficient tool for processing complex information, especially in the time required during sports games. This is especially true for team sports, where they must monitor a number of players at once.
Machine Learning can be applied to sports in a range of ways, with data now accessible about almost anything.
One example of machine learning being used in sport is in baseball. Ray Hensberger, a technologist in the Strategic Innovation Group at Booz Allen Hamilton, modelled MLB data that revealed what a pitcher would throw to a 74.5% degree of accuracy.
Hensberg and his team achieved this by looking at a pool of 900 pitchers over a three season period. They excluded players who had thrown less than 1,000 pitches total over the three seasons being examined, leaving an experimental sample of around 400, and then studied a range of metrics that could be used to determine what a pitch might throw. These included, among other things, the number of people on base, the hand used by the pitcher and the batter, the game situation, curveball release point, fastball velocity, general pitch selection, and slider movement.
The key to machine learning, according to Hensberg, is cross validation. Cross validation trains and tests machine learning models to ensure that they are not biased by the data they’re triangulated by, to provide insight on how the model will generalize to an unknown data set.
Wearable technology has also been a particular boon to machine learning. Algorithms are now being used to classify human movements, using recognition algorithms to identify individual athlete movement sequences in a variety of different sports using wearables.
As such technology improves, the amount of data being taken would become unmanageable were it not for Machine Learning, and coaches resistant to its use who fail to adapt may find themselves obsolete.