Technical–Tactical Determinants of Success Across the FIFA World Cup 2026: A Complete-Tournament Analysis Integrating Match Duration and Machine Learning

This study analysed the FIFA World Cup 2026 to identify performance indicators associated with match outcome, examine differences between group and knockout phases, and explore prediction using conventional and machine-learning approaches. All 104 matches were analysed, yielding 208 team appearances. Linear mixed-effects models evaluated 19 team-specific performance indicators with Match Result, Tournament Phase, Match Duration, and the Match Result × Tournament Phase interaction as fixed effects, while Match ID and Team were considered as crossed random intercepts and unsupported zero-variance components were removed. Match Result was significant for 15 of 19 team-specific indicators, with particularly strong evidence for goals, expected goals, attempts at goal on target, defensive pressures applied, and possession. No indicator showed a significant main effect of Tournament Phase in the primary linear mixed-effects models. The Match Result × Tournament Phase interaction observed for Goals in the primary Gaussian mixed model was not reproduced by the Poisson sensitivity model and was therefore not interpreted as robust. Total distance covered did not reach the 0.05 threshold for Match Result, F(2, 100.50) = 3.08, p = 0.0502, and Zone 4 distance was non-significant, F(2, 98.64) = 0.94, p = 0.396. Logistic regression showed discrimination (AUC = 0.816; accuracy = 0.736), with possession (OR = 1.041, p = 0.019) and attempts at goal on target (OR = 1.428, p < 0.001) independently associated with winning; mixed-effects sensitivity models estimated zero random-intercept variance for Match ID and Team. Two exploratory Random Forest classifiers were tuned exclusively within 83 training matches using repeated grouped five-fold cross-validation (10 repeats) and evaluated once on an untouched 21-match holdout set. The two-predictor model achieved holdout accuracy = 0.786, AUC = 0.793, and MCC = 0.520; the expanded 15-predictor model achieved accuracy = 0.714, AUC = 0.773, and MCC = 0.316. In the exploratory feature-importance analysis, attempts at goal on target and defensive pressures applied were the most prominent expanded-model features, whereas total distance covered had negligible permutation importance. Overall, success was associated primarily with attacking efficiency, possession control, passing, line-breaking, and final-third involvement rather than total running volume, while no consistent phase-related performance differences were detected after accounting for match duration.