# Data-driven classification of playing styles and match outcome prediction in UEFA Champions League teams

## TL;DR

This paper introduces a data-driven method to classify football teams by playing style and predict match outcomes using performance indicators.

## Contribution

A novel framework combining autoencoders, clustering, and machine learning models to classify and predict match outcomes for UEFA Champions League teams.

## Key findings

- SVM outperformed other models in predicting outcomes for possession-based teams.
- KNN was most effective for counterattacking teams.
- Defensive actions like clearances were strongly linked to match success in counterattacking teams.

## Abstract

This study proposes a data-driven framework for classifying UEFA Champions League teams into possession-based and counterattacking styles and predicting match outcomes based on key performance indicators (KPIs). Dimensionality reduction via an autoencoder was combined with K-means clustering to identify underlying tactical patterns beyond traditional possession metrics. Feature selection was performed using LASSO, Boruta, and XGBoost to determine the most relevant KPIs. Predictive models, including Support Vector Machine (SVM), K-Nearest Neighbors (KNN), and LightGBM, were evaluated using AUC and F1 Score. SVM achieved the highest performance for possession-based teams, whereas KNN outperformed other models for counterattacking teams. The results revealed distinct style-specific performance profiles. For possession-based teams, higher possession and key passes correlated negatively with winning probability, while crosses and long-range shots were positively associated with success. In counterattacking teams, increased possession and key passes improved match outcomes, whereas crosses and shots from outside the box showed negative associations. Defensive actions, particularly clearances, were strongly associated with improved defensive stability and match success, especially among counterattacking teams. This framework improves the accuracy of tactical classification and provides interpretable associations between KPIs and match outcomes. The findings can inform style-specific tactical planning and performance monitoring, enabling coaches to adjust offensive or defensive training priorities according to team strategy.