The growth of esports as a widely enjoyed activity for entertainment and social engagement has positioned it as one of the prominent domains of entertainment, gaming, and sports, attracting considerable interest from the research community. Considering that esports generate rich, real-time telemetry data, which can constitute the critical mass needed for researching data analysis and forecasting, a significant potential arises, especially in providing more engaging viewing and commentary. In this work, we study short-horizon event prediction in League of Legends, focusing on predicting imminent player elimination events (”deaths”) instead of overall outcome prediction during professional matches. We propose a time-series forecasting approach based on Temporal Fusion Transformers that uses multi-modal in-game data formulated as time-series to forecast near-term elimination events. Our system is designed to facilitate esports sportscasters, directors, producers, and content creators, attempting to surface likely high-impact moments and support proactive narrative cues. Trained in less than 200 matches, for a 5.0-second forecasting horizon and based on 10.0 s worth of historical data, our model achieves a score of around 0.6, which constitutes its performance comparable to similar work. This study demonstrates that short-horizon event forecasting is feasible using in-game data and transformer-based temporal models for the esports domain, thereby introducing a novel approach addressing a research topic that is rarely explored in esports analytics literature. Its findings suggest practical improvements for real-time support of esports’ storytelling experience and open avenues for future research on anticipatory analytics in interactive and spectator-driven digital sports.
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号