Project Abstract

I took on this project because champion synergy in League of Legends can get incredibly complex, especially as the game evolves with new updates and shifting metas. By using data-driven strategies, I can help players and analysts make smarter picks, ultimately pushing competitive play to new heights. My main goal was to create a Graph Neural Network-based system that looks at champion relationships in a more holistic way, tackling the question of which team compositions consistently achieve the highest win rates. This unique approach positions my work as a fresh take on esports analytics, leveraging massive amounts of game data in a graph-centric model. To make it happen, I built detailed champion relationship graphs from Riot API data using NetworkX, then fed those into a PyTorch pipeline for training and predictions. My system ended up with a notably higher accuracy in predicting win probabilities compared to traditional drafting methods, spotlighting the remarkable impact of synergy among roles and champion pairs. These insights also informed a pick recommendation engine, guiding players toward stronger drafts and reducing random guesswork. In the end, this project demonstrates how graph-structured data can unlock deeper layers of strategy and reveal previously hidden synergy patterns. Now, I have a clearer view of how to systematically boost win rates by combining champion attributes, historical performance, and a robust machine learning framework.

Keywords: Graph Neural Network, Predictive Modelling, Esports Analytics

 Conference Details

Session: A

Location: Sir Stanley Clarke Auditorium at 11:00 13:00

Markers: Cécilia Pradic, Adam Wyner

Course: BSc Computer Science 3yr FT

Future Plans: I’m looking for work