Sim2Win Predicts Football Outcomes and Tactics Without Team Identity

Pre-match tactical analysis in professional football often relies on subjective expert opinions and systems tied to specific team identities, limiting their generalizability to new or unseen teams. This research introduces Sim2Win, an event-based framework designed for pre-match tactical recommendations and outcome prediction that operates entirely without team names or identity features. Sim2Win processes StatsBomb event data from numerous competitions to create rolling tactical profiles for teams. It engineers four interpretable tactical feature ratios and clusters team behaviors into eight distinct playstyles using K-Means. Thirteen classifiers are then trained to estimate win, draw, and loss probabilities based on these tactical matchup representations. Rigorous evaluation using a Leave-One-Competition-Out (LOCO) method demonstrated that Sim2Win achieves a mean ROC-AUC of 0.704 and 55.4% accuracy on completely unseen teams, consistently outperforming traditional identity-dependent baselines like ELO and Pi-Rating. This indicates that behavioral tactical representations offer a robust and transferable predictive signal, even under distribution shifts.