Predicting female football outcomes by machine learning: behavioural analysis of goals as high stress events

The significant emergence of women’s football has stimulated considerable scientific interest, particularly in enhancing performance and achieving success. Football’s dynamic nature with its complex interactions and contextual variables, significantly influences player performance that can affect match outcomes. While goals are vital for securing a win, they can also trigger unexpected psychological responses such as stress and pressure potentially altering player behaviour and impacting the match’s trajectory. Effectively predicting and managing these behavioural shifts is important to in-game regulation. This study aims to enhance the performance and in-game success in women’s football by developing machine learning (ML) models that predict match outcomes based on player and team behaviour following goals. We applied a comprehensive approach that integrates spatiotemporal and behavioural data during the transitional period following goals focusing on team dynamics, including chaotic and collective behavioural analysis with entropy and fractality, spatial area, movement trajectories, and locomotor patterns. Several well-established ML models and feature extraction techniques were deployed with overall good performance of greater than 70% accuracy, with some specific methodology combinations have superior performance. Self-reported player wellness did not contribute to the predictions. In conclusion, game outcomes can be predicted with reasonable accuracy based on player behaviour during a relatively small proportion of game time, although this time represents events of high stress and pressure.