Efficient talent identification in women’s football: A ranking-based approach for goal scoring analysis

Abstract

Individual goal-scoring analysis in women’s football faces severe class imbalance and limited scouting resources, where classification metrics alone do not capture operational efficiency. We analyzed 2,535 non-goalkeeper player-match observations from the 2023 FIFA Women’s World Cup (736 unique players) with 51 performance features, excluding match-outcome variables to emphasize individual actions. Using nested cross-validation, LightGBM captured 79.4% of goal-scoring observations within the top 20% of ranked observations; an out-of-bag (OOB) bootstrap gains analysis yielded 73.9% capture at Top 20% (lift = 3.69x; 95% CI: 63.9%−84.3%). Permutation and SHAP consensus highlighted tactical availability (Total Offers) and combined technical/physical workload indicators (Passes Attempted, Jogging Distance, Top Speed). This proof-of-concept study shows that ranking-based evaluation improves scouting efficiency using basic match statistics, while thresholds and feature weights require validation in other competitive contexts.