A data- and knowledge-driven framework for developing machine learning models to predict soccer match outcomes

Aug 3, 2026

Upload by Violet

The 2023 Soccer Prediction Challenge asked the machine learning community to predict 736 future soccer matches using two tasks: exact score forecasting and outcome prediction as a probability vector over home win, draw, and away win. The study introduces a data- and knowledge-driven framework for building models from readily available data, centering on an innovative method for modeling interdependent time series of competing teams. Models using k-nearest neighbors, neural networks, naive Bayes, and ordinal forests were evaluated, with k-NN and neural networks achieving top performance, showing that simple algorithms can excel when domain knowledge is integrated well.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages40

Reading Time1 h 41 min

AccessFree to download

Introduction

Predicting soccer match outcomes and challenges

The data- and knowledge-driven framework

Tasks and evaluation in the 2023 Soccer Prediction Challenge

A data‑ and knowledge‑driven framework for developing machine learning models to predict soccer match outcomes Daniel Berrar1,2 · Philippe Lopes3 · Werner Dubitzky4 Received: 23 October 2023 / Revised: 22 July 2024 / Accepted: 3 September 2024 / Published online: 24 September 2024 © The Author(s) 2024 Abstract The 2023 Soccer Prediction Challenge invited the machine learning community to develop innovative methods to predict the outcomes of 736 future soccer matches. The Challenge included two tasks. Task 1 was to forecast the exact match score, i.e., the number of goals scored by each team. Task 2 was to predict the match outcome as probability vector over the three possible result categories: victory of the home team, draw, and victory of the away team. Here, we present a new data- and knowledge-driven framework for building machine learning models from readily available data to predict soccer match outcomes. A key component of this framework is an innovative approach to modeling interdependent time series data of competing entities. Using this framework, we developed various predictive models based on k-nearest neighbors, artificial neural networks, naive Bayes, and ordinal forests, which we applied to the two tasks of the 2023 Soccer Prediction Challenge. Among all submissions to the Challenge, our machine learning models based on k-nearest neighbors and neural networks achieved top performances. Our main insights from the Challenge are that relatively simple learning algorithms perform remarkably well compared to more complex algorithms, and that the key to successful predictions lies in how well soccer domain knowledge can be incorporated in the modeling process. Keywords 2023 soccer prediction challenge · k-NN · Ordinal forests · Naive Bayes · Neural networks · Outcome prediction · Soccer analytics · Super league 1 Introduction Unlike individual sports, such as tennis or golf, team sports are characterized by a much higher degree of complexity due to the vast number of possible interactions between players, moves, tactics, and strategy. Thus, predicting the outcomes of team sports games is extremely difficult. Soccer, arguably one of the most popular team sports worldwide, is a multi-billion dollar business. The modern game of association football is governed by the rules set forth by the Football Association Board and organized by bodies like FIFA (the Fédération Internationale de Football Association) and various continental and national Editor: Ulf Brefeld. Extended author information available on the last page of the article federations. Predicting the outcome of soccer matches has been a subject of research since at least the late 1960 s (Reep & Benjamin, 1968 ; Hill, 1974 ; Maher, 1982 ; Dixon & Coles, 1997 ; Angelini & De Angelis, 2017) . Over recent years, soccer match outcome prediction has gained increased attention from the machine learning community (Berraret al., 2019a) . The beauty of soccer match outcome prediction is that the fundamental task can be understood by practically anyone, and it is therefore also an excellent vehicle to showcase machine learning research to a wider audience. At the same time, it provides a truly exciting challenge for machine learning. While match outcome predictions are of interest to clubs, soccer associations, sports equipment and services companies, etc., it is certainly also incentivized by the betting industry (Malamatinos et al., 2022) . Part of the fascination of soccer is explained by the difficulty to predict the outcome of a match. If we viewed a soccer match as a scientific experiment to determine which team is better, we would realize that the number of robust measurements is rather limited. A soccer match involves hundreds of skillful moves and a wide variety of strategic and tactical plans, but the outcome is typically decided by a handful of quick and often random events, for example, a free kick, a mistake by a defender or goal keeper,

FAQs

What were the two tasks in the 2023 Soccer Prediction Challenge?

Task 1 predicted the exact match score by forecasting the number of goals for each team. Task 2 predicted the match outcome as a probability vector across home win, draw, and away win.

What is the core idea of the proposed framework?

The framework builds machine learning models from readily available data and emphasizes modeling interdependent time series for competing entities, i.e., soccer teams over time.

Which models performed best in the challenge and why?

Among submissions, models based on k-nearest neighbors and neural networks achieved top performances. The study highlights that incorporating soccer domain knowledge and using relatively simple algorithms can outperform more complex approaches.

Aug 3, 2026

Upload by Violet