Digital Creativity Labs develops real-time prediction model for esports

Researchers from Digital Creativity Labs at the University of York have announced the development of a landmark real-time prediction model for esports.

In a release received exclusively by Esports Insider, the team revealed the research of a model which “can reasonably predict the outcome of Dota 2,” and which could potentially alter the future of the esports broadcasting experience.

DC Labs' new model predicts Dota 2 matches with 85 percent accuracy
The Weavr companion match prediction AI in action in a game between Gambit Esports and TNC Predator.

RELATED: Dr. Florian Block – University of York – Researching esports

The model, which forms part of an advanced artificial intelligence and data infrastructure developed between DC Labs and its partners, is capable of predicting the winner of a Dota 2 game with an incredible 85 percent accuracy rate after just five minutes of play. Researchers are also able to reliably anticipate (within a five-second window) when a player character will get killed, predict with a good degree of accuracy the outcome of a match prior to the game commencing, and provide retrospective post-match analysis.

According to the release, the research has the potential to transform the way that spectators view and interact with esports by “enhancing audience understanding across all ability levels.” Unlike metrics which track in-game currency or objective control, which can sometimes be misleading or complicated to newcomers, the model’s statistics could be used by esports broadcasters to accurately tell the unfolding story of the match in a noob-friendly way.

For an esports title such as Dota 2, which is notorious for having a high barrier to entry due to its steep learning curve, the research may be game-changing.

RELATED: University of York, ESL to introduce esports content production module

The technology is at the heart of the Weavr project (led by ESL UK), with the option to stream to any touchpoint including mobile application, VR, and AR broadcast. DC Labs built the model using raw data from Dota 2’s millions of daily recorded matches – which it describes as “a unique and fertile testing ground for machine learning” – and trialled and evaluated the model during two Dota 2 tournaments.

DC Labs’ Dr Victoria Hodge said: “Esports, as with all sports, need to maintain an element of doubt to be enjoyable. However, Dota 2 games are overwhelmingly complex, even for experienced players and fans. For many esports viewers, the number and meaning of the statistics displayed can be confusing. It is often difficult to tell who is leading as the statistics can be contradictory.”

The model aims to provide a spectator experience that is “accessible and immersive,” and its ramifications could even be felt outside of the esports industry, as Dr Hodge explains. “In the future, when similar high frequency and detailed datasets are available from domains such as the Internet of Things, we can start to apply our live prediction to human behavioural data in the real world.”

The paper, ‘Win Prediction in Multi-Player Esports: Live Professional Match Prediction’ was published in the IEEE Transactions on Games in November 2019.

Esports Insider says: An 85 percent accuracy rate at five minutes for a game with as many variables as Dota 2 is astounding. This groundbreaking research could have significant applications within esports broadcasting, live betting, and audience experience.

Read The Esports Journal