Data-Driven Football Scouting System: How to Use Machine Learning to Discover Undervalued Players

In the modern football transfer market, clubs face the challenge of information asymmetry. The open-source project "data-driven-football-scouting" provides a systematic solution through a machine learning pipeline, helping to discover undervalued players, find realistic recruitment alternatives, and support multiple scouting workflows.

Football scouts need to answer five core questions: player market value, identification of undervalued players, alternatives for target players, reasons for priority review, and validation of the value of historical clues. The project started with a leak-proof market value prediction model, ensuring that only match data before the valuation date is used.

Data Layer: Integrates data from the Transfermarkt platform and advanced statistical data from the top five leagues; Model Layer: Two complementary XGBoost models (performance model for scouting, market-aware model for benchmarking); Similarity Engine: Role-aware similarity search combined with tactical adaptation scoring; Visualization Layer: Streamlit interactive dashboard supporting multi-dimensional filtering.

The project went through seven stages: 1. Market value prediction (leak-proof model); 2. Scouting dashboard (interactive tool); 3. Player similarity search (alternative matching); 4. Enrichment of advanced statistical data (position-specific profiles); 5. Role-aware similarity (tactical realism); 6. Scouting reasoning explanation (user-friendly interpretation and action recommendations); 7. Temporal validation (evaluation of historical clue value).

The system supports four workflows: 1. Identification of undervalued players (screening based on performance models); 2. Alternative comparison (ranking by statistical similarity and tactical adaptation); 3. Tactical compatibility assessment (role and position fit); 4. Historical validation and signal auditing (retrospective check of clue effectiveness).

For small and medium-sized clubs: Improves scouting efficiency and accuracy, discovering overlooked potential players; For large clubs: Provides alternative analysis and tactical adaptation data support. The system has transformed from a static tool to a decision-support workflow, helping scouts understand "why to focus" and the next steps.

The architecture is reusable and extensible, with data pipelines supporting multi-source data integration; The dual-model strategy (performance/market-aware) provides scenario flexibility; Role-aware similarity can be migrated to other tactical scenarios; The code structure is clear and documentation is complete, providing a reference for sports analysis projects.

This project demonstrates the potential of machine learning in sports analysis and is a complete solution for understanding and supporting business workflows. In the future, the combination of statistical modeling, tactical understanding, and business insights will become a standard configuration for scouting.