rare-event

RareEvent-Football: Team-Window Dataset for Tactical Rare-Event Analytics Short Description A football analytics dataset derived from StatsBomb Open Data for studying rare tactical and match-context events at the team-window level. Each row represents a sequential window in a team-match event stream, with engineered passing, tempo, disruption, and temporal-context features designed for rare-event detection and anticipation. About This Dataset This dataset was prepared for academic research on rare-event modelling in football. The raw event data was transformed into team-specific sequential windows, where each window summarizes tactical and match-context signals such as passing volume, pass success, forward progression, long passing, error clusters, substitutions, injuries, goals, cards, formation changes, and temporal shifts from previous windows. The main objective is to support research questions such as: Which tactical patterns appear before rare football events? Can a model anticipate a rare event within the next few team-windows? Which features are most useful for coach-oriented alert systems? How do rare-event windows differ from normal game-state windows? The final modelling version includes lagged, rolling, cumulative, and interaction features. It also includes leakage-safe train, validation, and test splits based on match_id, so researchers can evaluate models without mixing rows from the same match across splits. Dataset Scope Rows: 63,603 team-windows Matches: 868 Teams: 41 Main file columns: 100 Current-window rare-event windows: 12,364 Current-window rare-event rate: 19.44% Source: StatsBomb Open Data Unit of analysis: one team in one sequential event window Rare Event Definition A rare event is an engineered label that marks unusual or tactically important football moments. The current-window label is stored in rare_event, while rare_event_reason and rare_event_labels describe why a window was flagged. Rare-event triggers include: Red cards Second-yellow cards Penalty goals Own goals Low-xG or unexpected goals Substitution clusters Formation/tactical changes Error clusters, including errors, miscontrols, and dispossessions Recommended Files rare_event_modeling_dataset_enhanced_plus.csv The main research-ready dataset. Use this file for modelling and analysis. Important columns: match_id: match identifier team: team name time_window: sequential team-window index within a match rare_event: whether the current window contains a rare event rare_event_reason: human-readable explanation of the current-window rare event target_next_rare_event: whether the next team-window contains a rare event target_next_3_windows_rare_event: whether a rare event occurs within the next three team-windows split: leakage-safe split label: train, validation, or test lag/rolling/cumulative features: temporal context for rare-event anticipation final_model_predictions_test_split.csv