Winning With Chaos in Association Football: Spatiotemporal Event Distribution Randomness Metric for Team Performance Evaluation

Association football (commonly known as football or soccer) in the modern era places a greater emphasis on collaborating and working together as a team instead of relying solely on individual skills to strategize winning performances. The low-scoring and unpredictable nature of association football makes evaluating team performances challenging. Space creation and space utilization have been discussed in the football world lately. Existing literature evaluates this with on and off-ball runs by players for deceiving defenders to create open spaces. However, the contribution of these team ball movements’ enhanced randomness or chaotic nature to winning performances has yet to be explored. This work proposes a novel entropy-based time-series performance evaluation metric, EDRan, for quantifying this enhanced random nature by analyzing the spatial distribution of game events at regular intervals. Additionally, an unexplored cumulative ball possession matrix is used to quantify randomness. The correlation between the match winner and spatial event distribution randomness at regular intervals was analyzed. The significance of the proposed metric was demonstrated using a generalized linear model (GLM), which achieved an average accuracy of 80% for match-winning performance classification. The GLM p-values and coefficients revealed statistically significant relationships between the extracted temporal features and match-winning performances. Findings further revealed dispersed, highly random event distribution by winning teams during the early phases of the game, implying attacking behavior, followed by a compact, cautious playing style toward the end, suggesting that the game’s first-half performances are more pivotal. Despite the unpredictability of actual scores in association football, the proposed approach effectively captured the differences in performances between stronger and weaker teams with temporal relationships, highlighting its significance as a time-series metric for performance evaluation.