This eSports analytical model can predict which team will win

Researchers model new way to evaluate player performance and impact on a match.

While math might not be on the minds of most gamers while they’re holding down mid lane or saving their ADC’s life (again), eSports analytics is important to both the professional scene and game developers. This month a group of researchers presented an analytical model that not only evaluates player performance, but can predict which team will win the game.

The researchers analyzed 412 DOTA 2 games, dividing them into encounters between players, and collected data such on who won the fight and rate of experience gain. One of their experiments examined the likelihood that the team having more gold after a certain amount of time would win the match. Turns out, the team that has earned more gold after the first 5 minutes of the game has a 70% chance of victory. That shoots up to 90% after 15 minutes. The research also shows that 82% of experience and 79% of gold gain are earned in fights between players, and how much of an impact a particular player’s actions had on a victory or defeat.

This model could be an additional tool for developers to visualize how their game plays, or to use when designing updates. Its value to professional players is obvious, but non-professional players would enjoy a program that could analyze replays of their games and give tips on where they could improve. Although the team used DOTA 2 for their initial research, they hope to investigate its usage in other MOBAs such as League of Legends or SMITE in the future.


Teevell is a professional writer and editor, and is primarily interested in MOBAs, MMORPGs, eSports, and game writing.

Published Mar. 20th 2016

New Cache - article_comments_article_36002