How data proved key to Benfica’s title winning season

While skill, hard work, discipline and a will to win were key to the club’s most recent triumph, there was another factor employed that has also proved pivotal – technology, and in particular, data modelling.

The Portuguese Primeira Liga is small in comparison to the other major leagues in Europe (England, Germany, Spain and France), and as such does not have the financial clout of its bigger neighbours. This has seen the club adopt a cost effective model for developing young talent which it can then sell to Europe’s major clubs – creating an impressive revenue stream.

However, a vital component of this model is ensuring that the club’s youth players can develop injury-free. It’s here that data modelling is being employed. At Benfica’s Caixa Futebol Campus, where over 100 players train each day, every youth player is closely monitored on parameters such as sleep patterns, diet, speed, fatigue, stress, recovery and mental state. This data is then analysed in the club’s lab.

Using Microsoft Azure machine learning, this raw player data can then be used to optimise the players in preparation for match days, personalise training sessions and predict and prevent injury. Providing data is one thing, but making it actionable is quite another, but machine learning is making this a much easier task by storing team and player data into what the club call a single ‘data lake’, which staff can then dip into to predict future injuries, recovery and performance levels.

This is achieved by using a custom middleware layer that collates the output from each disparate data system that the club uses into a single format. This is then stored in the data lake on the club’s own dedicated data server. Security is also a key concern for the club, and therefore access to the data is segregated to protect confidentiality.

While the club is still at the start of the machine learning journey, its ultimate goal is to develop a highly accurate injury prediction model which will not only be able to predict when a player may become injured but also the training threshold (i.e. how far can they push themselves before getting injured), the type of injury and the length of recovery time.