Riot Games is teaming up with Amazon to create Win Probability AI to predict LoL matches for Worlds
Machine learning and AI has been all the craze in the tech world as of late, and now it's finally sophisticated enough to predict League of Legends matches.
As the latest development in the LoL esports scene, Amazon Web Services (or “AWS”) officially announced it's collaboration with Riot Games and the formation of it's new machine-learning (ML) model for League of Legends. Aptly named “Win Probability”, Riot's shiny new AI takes factors like Tower Kills, Inhibitor Timers and even Total Team EXP into account to figure out which team has the best chances of coming out on top.
The tool makes for an awesome supplement that simplifies all of the mathematical mumbo-jumbo for newcomers, and makes a playcaster talent's role smoother than ever before. Here's everything you need to know about Win Probability for Worlds 2023.
What is Win Probability?
Win Probability is a “machine learning-based tool” that suggests which team has a high likelihood of winning when compared to past scenarios/games. Win Probability does this by generating a percentage based on a multitude of in-game metrics via xgboost, a versatile machine-learning algorithm used for user-defined ranking and predictions.
Win Probability was developed by Amazon Web Services (AWS), an Amazon subsidiary that specializes in cloud computing and APIs. AWS along with Riot Games decided to debut the tool during the Mid-Season Invitationals back in May this year, but back then it was called “Win Expectancy (WE)”.
AWS also published an insightful look into Win Probability's development process on their website after Worlds kicked off this year. To know more about the modular workflow that resulted in Riot's latest ML analytics tool, click here.
Spectators in the 2023 Worlds Championship will be able to see the Win Probability tool in action.
Which Factors Does Win Probability Use?
In the official LoL Esports dev diary, John “Riot JPham” Pham confirmed that Win Probability recorded and analyzed more than 3 years worth of data from major LoL esports matches to create its sophisticated prediction model. These are all of the factors AWS and Win Probability take into consideration:
- Match time
- Gold % (player gold / total gold in game)
- A team's total in-game experience
- Number of players alive
- Tower score
- Dragon souls
- Secured Herald trinkets
- Inhibitor respawn timers
- Baron buff timers
- Elder Dragon buff timer
- Number of living players with Baron active
- Number of living players with Elder active
Win Probability was trained using data from professional LoL esports matches since patch 10.4 (around February 20, 2020). Win/loss trends as well as every succeeding game helps Win Probability account for any sudden changes in the meta.
An Exciting Addition to LoL Esports
As exciting as the prospect of analytics prediction sounds, the devs recognized that the new sports data model is still in it's early stages. They're starting with the broad strokes, and believe that there's more to come, especially since big key factors like champion composition and item changes aren't being included in Win Probability's algorithm. With that in mind, Riot Games hinted at some important developments down the line, including WP's next evolution.
Win Probability Added, or WPA, will be the machine-learning tool's ambitious next step. Considering not only those previously mentioned but specific actions the players take, the next big iteration could determine “which players are making the most impact”, Pharm remarked, “and maybe even answer the age-old question of “Is Faker the GOAT?” once-and-for-all.”
The blogpost went on to highlight the company's commitments surrounding the use of their new AI tools for Worlds 2023 and beyond, namely:
- Refreshing the WP graphic to be more visually distinct
- Increasing transparency about the underlying technology
- Improving education for our fans, casters and community
For the latest news on esports, follow us on ESTNN.