Assessing player performance in tactical FPS games: A proposal for an xGoals-inspired model
Pete Gordon, William Kavanagh, T. A. Howson, David E. Lavallee · 2024
As the professional esports industry continues to grow, so too does the demand for sophisticated performance analysis tools. This paper proposes the development of a machine learning model called ‘xKills’ to quantitatively assess player performance in first-person shooter games (inspired by the ‘xGoals’ concept from football analytics). Our focus thus far has been creating the necessary software tools to: a) simulate spatial context for telemetric replays of professional matches in a popular esports first-person shooter game, b) analyse gameplay to identify scoring opportunities, and c) extract feature sets that describe these opportunities and their outcomes as training data. Preliminary validation of these tools confirms their efficacy in capturing detailed features that influence scoring probabilities. Although the model has not yet been trained, the groundwork laid by these tools is crucial for the forthcoming steps, which will involve supervised training via a cross-validation approach. Ultimately, the xKills model aims to enhance the esports ecosystem by enabling more effective strategy development for professional teams and enriching broadcaster analysis.