Uncovering Global Football Scoring Structures Through Symbolic Regression and Similarity Networks

This dataset contains the processed shot-event data, engineered features, and analysis outputs supporting the manuscript "Uncovering Global Football Scoring Structures Through Symbolic Regression and Similarity Networks" (Cengiz, M.A., Ebrahim, E.A., Alharthi, A., & Kara, M.), currently under review at Expert Systems with Applications. The study applies symbolic regression (PySR) and similarity-network analysis to shot-level event data from ten major football competitions to uncover interpretable mathematical structures underlying goal-scoring opportunities. Contents Copa_America_raw_shots — 790 raw shot events from 32 Copa América matches, retrieved via the statsbombpy API (competition_id = 223, season_id = 282). All original StatsBomb event fields are included; nested fields (location, freeze frame, tactics, etc.) have been flattened to plain text for spreadsheet compatibility. This subsample was used for the multicollinearity (VIF) diagnostic reported in the manuscript's Supplementary Material. Engineered_features — The 40-column engineered feature table (shot geometry, goal-frame geometry, tactical context, temporal, and interaction features) derived from the shot events above, as described in Section 3.3 of the manuscript. Original_final_results — Model comparison results (ROC-AUC, PR-AUC, Brier score, Expected Calibration Error, runtime) across all ten competitions and all benchmark/proposed models. Original_similarity_matrix — The pairwise symbolic similarity matrix across the ten competitions used to construct the similarity network and hierarchical taxonomy (Section 4.7–4.8). A README sheet with column-level descriptions and provenance notes is included in the workbook. Data source and attribution The underlying shot-event data were obtained from StatsBomb Open Data (https://github.com/statsbomb/open-data), made freely available by StatsBomb for research and non-commercial analytical use. In accordance with the StatsBomb Open Data usage terms, any research, analysis, or derivative work based on this data should cite StatsBomb as the original data source. This deposit provides the authors' own derived features, model outputs, and analysis results; it is intended as a companion resource to the associated manuscript and does not alter the terms under which the original StatsBomb data are made available.