Predict League of Legends Game Outcomes using SAP Analytics Cloud
Since 2018, SAP has set up a partnership with Team Liquid, a professional esports organization, and helps them analyze their performance and create new strategies.
The goal of the partnership between SAP and Team Liquid is to co-develop software based on available (in-game) data which will help Team Liquid analyze performance and apply more precision in areas like team and player performance, scouting and others. By applying innovative technology to improve the quality of competitive play – as well as scouting future talent – and learning and growing with Team Liquid, it is SAP’s goal to contribute meaningfully to the esports ecosystem and to the experience of fans across the world.
The game is a constant switch between action and reaction by both teams. Though the number of possible actions is way too big to practice all of them. There are some scenarios that need warrant analysis, but even those have so many combinations so only a smaller and more relevant subset can be trained.
This blog post explains how we helped Team Liquid use SAP Analytics Cloud to predict game outcomes.
Introducing League of Legends
League of Legends is a multiplayer online battle arena strategy video game. Two teams of five players (red and blue) fight over a map. Each team defends their area while attacking the opponent. Each player controls a character (known as a “champion”). The goal is to destroy the opponent’s base (the Nexus).
Champions have different abilities, which are instrumental in the game. Players are positioned in three lanes and the jungle area. The five possible player roles include top, mid, bottom, jungle, and support.
The game starts with a champion selection phase called “the draft.” This is a strategic phase to prepare the game. The draft consists of 4 phases where teams pick their champions and ban the opposing team from picking their favorite champions.
A strategy example consists in banning a champion experienced in mid-lane to have better control of the mid lane. Another example: A player experienced in bottom lane will pick a champion experienced in bottom lane.
Phase 1: Three champions are banned by both teams.
Phase 2: Three champions are picked by both teams.
Phase 3: Two champions are banned by both teams.
Phase 4: Two champions are picked by both teams.
Multiple factors can help predict the result of the game:
- Historical team win rate (%)
- How players are experienced with champions and roles
- Champion and player win rates in both teams and so on.
End-to-End Data Flow
The data used relates to professionally played games. The diagram above presents the end-to-end data flow and transformation.
Team Liquid receives the data sets of all professional games from the esports data supplier “Bayes Esports” who collects all the records directly from Riot Games, the publisher of League of Legends. These data sets are ingested in SAP HANA Cloud. Python code allows to engineer extra features and create a CSV file ready to be imported into SAP Analytics Cloud. Next, we build a predictive model that present the results in a user-friendly way using the functionality of stories and dashboards.
SAP Analytics Cloud Analysis Results
We gathered every professional game played between July and September 2022. Using SAP Analytics Cloud, we created a predictive model. In the dataset, there are various features like team win rate, champion mastery for every lane, and player experience for every lane.
Our model can predict the result of the game before the draft takes place and after the draft takes place. Team Liquid can use these predictions to have prior knowledge about the game factors which influence the probability of winning. They can detect patterns, strengths, and weaknesses of competitors and finally, define their game strategy best.
Using an SAP Analytics Cloud story, we can present the predictions, and we can see what the predictive model is doing in a more comprehensive way. We used a table, as it is a straightforward way to summarize the necessary information.
On top of predicted win probability, one wants to understand the business reasons behind the win probability. What are the top reasons and how much do these reasons influence the model predictions?
The story shows the reasons for the prediction in a bar chart. In the right side, the chart shows the reasons that are in favor of the red team and the left side is for the reasons in favor of the blue team.
Let’s take an example:
The first line of the chart is named “team_win_rate_difference.” It means that the red team has a higher win rate, in the past they have won more games than the blue team. So, this is an advantage for them. The higher the bar, the greater the influence on the prediction.
To see the importance of the draft part, let’s take another example. Before the draft, the model predicts that the red team will win the game with 63% chance. There are multiple reasons in favor of the red team. For example, in support position and in mid lane their players have a good win rate. (5th and 6th lines)
Before The Draft Takes Place
After the draft, the probability of winning for the red team has decreased. Blue team had made the right draft choices which helped them win. The prediction explanations charts help understanding the why behind the prediction.
After The Draft Takes Place
In conclusion, we saw the model predictions for the game results and the influencers of these predictions. This is a way to help Team Liquid using SAP Analytics Cloud to improve their strategies.
Thanks for reading this blog, I hope you liked it. You can read more ‘Smart Predict’ blogs through this link: https://blogs.sap.com/tag/smart-predict/