Deep learning techniques for identifying KPIs in League of Legends: Win prediction, map navigation, and vision control

Data analytics are essential for improving team performance and strategy development in the rapidly changing environment of computer gaming and esports. League of Legends (LoL) is one of the most popular Multiplayer Online Battle Arena (MOBA) games and has greatly contributed to the rise of professional esports. In this study, we applied deep learning techniques to predict LoL match results and identify Key Performance Indicators (KPIs) that influence win probability at different stages of the game. We applied neural network models to a dataset provided from Riot Games' official API comprising of 154,000 games and achieved 97 % accuracy, demonstrating that the models effectively captured the critical factors influencing match outcome. The results confirm that reducing the number of turrets lost is the most important metric, followed by increasing bounty level and the number of turrets destroyed, reducing damage taken, and increasing the number of dragons killed. We then expanded our KPI analysis by investigating player movement patterns and map awareness. Our findings show that higher-skilled players spend more time near key map objectives, which can be verified by effective ward placement and overall vision control. This analysis deepens our understanding of how players and teams navigate the game map, manage vision, and factors that influence their performance and match outcomes. This approach provides practical insights for game developers, coaches, and players to help them refine strategies, improve competitive play and optimise esport performance.