Inspiration
Football has always been a game of decisions. Behind every match, coaches, analysts, and scouting teams have to answer hundreds of questions:
- Why did we win or lose?
- Are we improving or declining?
- Which players fit our tactical identity?
- What are the real factors behind the next opponent?
- Are we making decisions based on data or intuition alone? With modern football reaching unprecedented levels of tactical and analytical complexity, we see more than ever how much teams rely on data-driven decision; they need to understand thousands of data points and transform them into clear decisions under pressure. However, football data is everywhere, but meaningful insights are not. Most analytics tools provide numbers, charts, and statistics but a coach does not need another spreadsheet. A coach needs answers. This inspired us to build CoachIQ : an explainable football intelligence platform that transforms raw football data into coach-ready insights.
What it does
CoachIQ combines machine learning, football analytics, and explainable AI to answer three major football questions:
- How strong is a team and how is it evolving? Our Team Intelligence module analyzes historical performances, Elo strength trajectories, attacking and defensive profiles, and tactical characteristics to generate a clear understanding of a team's identity.
- Which players match a specific playing style? Our Player Intelligence module acts as a scouting assistant. Instead of simply searching for famous players, it finds footballers with similar playing profiles using machine learning techniques such as similarity search, PCA, and clustering. For example, a scout can discover players who have a similar style to a star player but are younger or less known.
- What are the chances of winning the next match? Our Match Prediction module uses explainable machine learning to estimate Home/Draw/Away probabilities while showing the football factors behind the prediction. Unlike many prediction systems, we do not use bookmaker odds as an input. We build our predictions from football information only and compare our results against the market as a benchmark.
How we built it
CoachIQ was built as an end-to-end football analytics pipeline, transforming raw historical football data into explainable decisions. We started with the European Soccer Database, containing more than 25,000 matches, player profiles, team attributes, and historical performance data across 11 leagues. The first challenge was preparing reliable football data. We created a data processing pipeline that: Explored and cleaned the relational SQLite database Connected matches, teams, and players through their relationships Applied time-aware data processing to ensure that models only used information available before each match Built a feature store containing reusable football features for the application For Team Intelligence, we engineered football-specific indicators: Elo ratings to track team strength evolution over time Recent form calculations Home and away attacking/defensive statistics Tactical profiles from team attributes For Player Intelligence, we transformed FIFA player attributes into machine learning representations: Standardized player characteristics using feature scaling Used similarity algorithms to find players with comparable playing styles Applied clustering techniques to discover player archetypes such as creative wingers or playmakers Used PCA visualization to represent player styles in a 2D space For Match Prediction, we built an explainable machine learning pipeline: Created pre-match features based on team strength, form, attacking output, defensive ability, and lineup quality Trained classification models to predict Home Win / Draw / Away Win probabilities Used a temporal validation strategy by training on earlier seasons and testing on future matches Applied SHAP explainability to understand why the model made each prediction Compared our independent predictions against bookmaker probabilities without using betting odds as model inputs Finally, we developed an interactive application using Streamlit and Plotly, allowing users to explore teams, scout players, analyze matches, and understand model decisions through visual reports and natural-language insights. For tactical analysis, we integrated StatsBomb event data to showcase advanced football analytics such as expected goals (xG), shot maps, and passing networks.
Challenges we ran into
The biggest challenge was not building a model it was building a trustworthy one. Football data contains many hidden problems. Player and team attributes change over time, meaning that using future information would create unrealistic predictions. We solved this by implementing time-aware feature construction, ensuring that every prediction only uses information available before the match. Another challenge was balancing accuracy with explainability. A model that predicts correctly but cannot explain its decisions is not useful for coaches. Therefore, we focused on transparent predictions where every result comes with understandable football reasons. We also learned that great analytics is not about producing more numbers. It is about transforming complex data into decisions that humans can trust.
Accomplishments that we're proud of
Building a complete football intelligence platform from raw data to interactive insights Creating a prediction model that achieved 50.8% accuracy on future matches, coming within 1.4 percentage points of bookmaker predictions, while never using betting odds as input Designing an explainable AI approach where every prediction is accompanied by the football factors behind it Building a scouting system capable of finding players with similar tactical profiles rather than simply comparing popularity or overall rating Preventing common machine learning mistakes through temporal validation and avoiding future data leakage Combining different areas of football analytics: team performance, scouting, prediction, and tactical analysis into one unified platform The biggest accomplishment is turning complex football data into insights that a human decision-maker can understand and use.
What we learned
Building CoachIQ taught us that machine learning becomes truly valuable when it solves a real-world problem. Football is unpredictable, but data can help teams make better decisions. Our goal was not to replace coaches, but to give them an intelligent assistant that helps them see patterns, discover opportunities, and understand the game from a deeper perspective. CoachIQ is our attempt to bridge the gap between football knowledge and artificial intelligence — turning millions of data points into insights that can support the people who make decisions on the pitch.
What's next for CoachIQ
CoachIQ is designed as a foundation for a more complete football intelligence assistant. Future improvements include: Integrating real-time football data for live analysis Expanding tactical analysis using more event-level data Adding video analysis and computer vision for player movement analysis Improving scouting recommendations with market value and career trajectory Developing personalized dashboards for coaches, scouts, and analysts Adding more advanced football models such as goal prediction models and tactical simulations Our long-term vision is to create an AI assistant that helps football professionals make faster, smarter, and more informed decisions.
Log in or sign up for Devpost to join the conversation.