Key performance indicators of offensive transitions in elite women’s football: a machine learning and explainability approach

Women’s football has experienced significant growth in sporting, economic, and social interest. However, there remains a shortage of studies examining key technical-tactical performance indicators in elite competitions. This research aimed to identify and quantify the influence of technical-tactical indicators on the effectiveness of offensive dynamic transitions in elite women’s football, using a machine learning approach. To this end, 3,610 dynamic offensive transitions recorded across 35 matches from the final stages of the FIFA Women’s World Cup 2023, UEFA Women’s Euro, and UEFA Women’s Champions League 2023/24 were analysed using an observational methodology. An ad hoc observation instrument, Transfootb, was developed to record teams’ offensive behaviours, opponents’ defensive responses, and the match context at the time of the offensive dynamic transition. A chi-square test was applied to identify associations between variables, followed by the training of a Random Forest model to predict transition outcomes. Additionally, ShAP values were computed and visualised to interpret the influence of the predictors. The model achieved an area under the curve of 0.78, with a recall of 18% and a specificity of 96%. The results indicate that the execution mode of offensive transitions and the match context significantly influence offensive success. Specifically, penetrating passes (≥ 3), counterattacks, and an opponent’s low defensive positioning were the key predictors of successful offensive transitions. This study provides valuable scientific evidence to optimise strategic decision-making in dynamic offensive transitions in elite women’s football. Furthermore, it highlights the potential of machine learning in analysing and predicting performance in sport-specific actions.