IMG ARENA predicts golfer performance with Amazon SageMaker

In an era of advanced analytics, fans of professional sports expect more than subjective commentary as part of the viewing experience. They seek data-driven insights, which is where company IMG ARENA (IMGA) excels. Harnessing the power of official sports data to deliver fan-centric experiences, IMGA has helped rights holders, the media, and betting industries offer next-generation entertainment for more than a decade. For its latest advancement in sports insights, IMGA’s data science and product team worked with Amazon Web Services (AWS) to develop a predictive machine learning (ML) model for golf player performance using Amazon SageMaker.

“We like to maximize storytelling around sports and sporting events, and leaning on the AWS tech stack enables us to do that. We can capture data in real-time and feed it into our cloud in half a second. From there, AWS helps us get insights out to various consumers,” said Eireann Kelly, Vice President of Product at IMGA. “By leveraging ML and Artificial Intelligence (AI), we can step beyond captured data and derive insights from that data to tell deeper stories.”

Collaborating with the AWS Generative AI Innovation Center team over the course of a 12-week engagement, IMGA leveraged data from 1,240 professional golf tournaments held from 2003 to 2021 to train the model. Data feeds were provided by the PGA Tour and DP World Tour, and IMGA collected its own data from other tournaments held outside the US. Various data inputs included historical tournament results, player data, and skill-distance between players. Data inputs were stored in Amazon Simple Storage Service (Amazon S3) and fed to Amazon SageMaker to predict performance in future events. The team looked at player ranks in previous tournaments and player stats for the last 30, 120, and 365 days. For the predictive model, they fed all the player and tournament features, along with static ranks computed using the Massey’s Method equation for determining rankings, to the model. The model was then evaluated for accuracy against 65 tournaments held in 2022.

“In any sport, especially golf, you’ll have lucky days, so you need to appropriately measure those noisy fluctuations. Also, the course is always changing, unlike pitch sports. Our focus was establishing a solid baseline. We went through a lot of variations on what data to look at beyond number of strokes. There is so much data to look at that it’s impossible for a human to parse through it all, which is why ML is so valuable,” Kelly explained.

The team designed a system with an accuracy improvement of 4.5 percent when compared to industry standard. Given the model’s success, it will now go into production for internal use to enrich IMGA’s data products. Unlike golfer rankings, which don’t often change drastically, predicting performance is a dynamic value that changes with each stroke. With a proven foundation in place, the model can also evolve and scale to predict win probability for use in broadcast graphics, for fan engagement, and in other AI products, such as AI commentary. To that end, the IMGA team has started experimenting with Amazon Bedrock, an AWS service that enables customers to build and scale generative AI-based applications using foundation models from Amazon and leading AI startups, including AI21 Labs, Anthropic, and Stability AI, accessible via an API.

“Using ML enables us to unlock infinite variables for our storytelling, and that level of flexibility isn’t supported in traditional probability and rating work done with old school approaches,” concluded Kelly. “Now, we have the tools to harness unstructured data, and further elevate the fan experience with powerful applications using generative AI.”

Learn more about how to build train, and deploy ML models for any use case with fully managed infrastructure, tools, and workflows on AWS.