Inspiration
Having grown up playing league and watching the competitive esports scene, I wanted to bridge my current studies in Machine Learning and Computer Science and my love for the game. This Hackathon was a great way for me to do so!
What it does
This model looks to predict the winner of a league of legends game based on the game state info, but specifically in order to understand what variables are most important, and how they change between years. Understanding how impactful an early dragon might be, as compared to an early gold lead, can help pro players and coaches make the right call when it matters most.
How we built it
A large portion of this project was focused on data cleaning, which required some deal of background playing the game. I was able to take my years of experience in NA silver lobbies and try to develop meaningful variables from the dataset. The dataset is linked here for anyone's interest: https://www.kaggle.com/datasets/chuckephron/leagueoflegends
Challenges we ran into
The only main challenge was that League of Legends has a lot of champions. Including the champions in the dataset led necessitated some form of encoding, to which one hot encoding was selected. This did create a great deal more columns; however, considering the accuracy achieved, this was not a terribly negative impact.
Accomplishments that we're proud of
Identifying the slight shift towards an earlier game gold dependency from 2016 to 2017 is consistent with what many players experienced as Riot Games made the games shorter. Having the gold lead earlier on matter more is clearly indicative of a meta shift, which was amazing to see.
What we learned
Data cleaning is a super tedious process; however, it can also be very rewarding. This was the one area of the project where I was able to introduce my own knowledge about League of Legends to improve the model, which was exciting to experience.
What's next for Competitive Esports Prediction Model
Data augmentation. The lack of data is a major setback, and while the solo queue community could be examined, the difference between competitive esports and solo queue is a stark one. Data augmentation could allow for the amount of data needed to explore larger scale models, such as Deep Learning Neural Nets.
Log in or sign up for Devpost to join the conversation.