Analysis of data from the videogame/eSport League of Legends

Shown are results from 2014 World Championships to end of quarterfinals of 2019 World Championships. There may be some lapses due to lack of data; some of the teams are merged by name (notably, Gen.G refers to all Samsung organizations; ROX and KOO are equivalent).

Until halfway through Season 8, a Korean team was always the highest rated team (except for a tiny blip of TSM during Season 7): one of SKT, ROX, or Gen.G. In particular, SKT was the strongest team for almost three straight seasons. In Season 8, FW overtook Gen.G before being overtaken by IG, the eventual Season 8 World Champions. In Season 9, GRF, SKT, and IG all briefly held the top spot, until G2 held it for the latter half of the season. Now, nearing the end of the Season 9 World Championships (matches up to 11/03/19), G2, SKT, and FPX are all near the top.

These were the predictions given by the model for each of the playoffs matches:

Quarterfinals

• Griffin has 69.07% probability of beating Invictus Gaming
• G2 Esports has 56.68% probability of beating Damwon Gaming
• SK Telecom T1 has 74.42% probability of beating Splyce
• Funplus Phoenix has 50.54% probability of beating Fnatic

Semifinals

• Funplus Phoenix has a 64.36% probability of beating Invictus Gaming
• G2 Esports has a 50.01% probability of beating SK Telecom T1

Finals

• G2 Esports has a 54.14% probability of beating Funplus Phoenix

* at the end of a patch

10/27/19: (Initial commit) Shows elo ratings over time for various teams

This is an ongoing project that I will work on occasionally. Feel free to use the scripts/results for your purposes; please cite me (Kevin Lu).

See 538 for a more detailed description of Elo ratings.

I set K=20 for updates. To initialize teams faster, I use K=100 for the Season 4 World Championships. To emphasize importance of the world championships and to help blend ratings of different regions, I set K=40 for updates during a world championship. Some of the teams are merged by name (this may not be fully complete).

If you wish to run the scripts yourself, please download data from:

Oracle's Elixir

Chuck Ephron's Kaggle Dataset

Scripts are written in Python 3. Visualization is done with numpy, matplotlib, and seaborn. To set the text labels, I used adjustText. All of these can be installed with pip via the command line.

Python