How the National Football League uses AWS

Just in time for the start of the playoffs in the North American NFL, Amazon Web Services has launched a campaign to draw attention to its cooperation with the National Football League, which has been in place since 2017 and has been constantly evolving ever since, and has published various details. The centerpiece of the cooperation is the NFL software Next Gen Stats (NGS), which runs on Amazon Web Services and is responsible for countless developments within the NFL with the help of AI and cloud-based machine learning. For example, simulations in the cloud can be used to optimize team training and reduce the risk of injury. At the same time, the data from AWS, which is obtained by analyzing data points collected during games, is used to develop new, better equipment. But that is by no means all.

As the official technology provider of the NFL, AWS is involved in all phases of the development and implementation of Next Gen Stats. Data is collected via ID tags on the players' shoulder pads and on the ball, and all data is automatically sent to the AWS cloud. In this way, almost 300 million data points are generated each season during the games, which AWS stores and processes. Tools such as Amazon SageMaker are used to quickly develop and train ML models and provide the analyzed results. Another tool, Amazon QuickSight, is used to analyse and visualize the statistical data.

Challenges and solutions in machine learning

An interesting field of application for NGS is so-called "defender ghosting". This involves predicting the running paths of defenders as soon as the ball leaves the quarterback. These predictions are not only important for analyzing a quarterback's decision-making, but also for developing new plays. For example, it is possible to calculate how a play would have developed if the quarterback had passed the ball to another receiver.

With the help of Amazon SageMaker, the NFL's NGS team has also developed a machine learning model that looks at the probability of a pass successfully reaching the desired receiver. This model takes into account ten measurements on the field, such as the distance of a pass or the distances between the quarterback and the various opponents and his own teammates. It also includes probabilities from past plays as well as predictions about the flight behavior of the ball and the behavior of the defensive players. The latter changes constantly over the course of the game depending on the score and previous events.    

To develop extremely complicated models like this, the NFL team teamed up with various AWS scientists. They filtered anomalies from the training data and focused on the most important features for their predictive models. By combining NFL expertise and AWS expertise, they were able to develop precise models for Defender Ghosting, for example.

Another area in which AWS works for the NFL is the creation of the league's schedule. This used to be done manually, which took several months. Today, AWS simulates a trillion different combinations for the NFL and determines the perfect game plan in a fraction of the time.

The future of sports statistics

The results of the NFL and AWS collaboration allow for more detailed and accurate analysis of football games. The bottom line is that these advanced statistics not only provide insights into actual plays that took place, but also allow hypothetical scenarios to be evaluated. As a result, player decisions can be better understood and analyzed.

So NGS, supported by AWS, opens a new chapter in sports statistics that not only changes the way games are analyzed, but can also enrich discussions among fans and experts. In addition, technological advances are helping to deepen understanding of the game and provide new perspectives on player performance.

Solving business problems with "The Machine Learning Pipeline on AWS"

Learn in the course "The Machine Learning Pipeline on AWS"how you can use Amazon SageMaker to implement specific projects to solve business problems such as fraud detection or recommendation systems. Over four days and eight modules, you will learn how to create, train, evaluate, optimize and, of course, implement machine leaning models with Amazon SageMaker.