Football Match Results Data Set for Prediction (Machine Learning)

You can download the Data Set and the project code by clicking HERE.

With this unusual post, we are kicking off a new initiative within the context of Machine Learning (the field of study that aims to give computers the ability to learn without being explicitly programmed) and we hope it will be successful and get a following. This initiative is the publication of a Data Set of football match results from the Spanish first and second divisions, from the 1970-71 season to the current one. We are sharing code written in Python to "scrape" the data and keep this data set up to date, and depending on the interest this initiative gets, we will keep updating and maintaining the code to obtain more football results.

The goal of publishing this Data Set is to open a new area of research within Machine Learning for predicting football match results; at first, producing a result in pools coupon format (1X2), although it could be extended to predicting the final score.

The motivation for this initiative comes from the final project I did for the course "Neural Networks and Applications" taught in the "Master in Computer Science and Technology" at the Universidad Politécnica de Madrid (UPM), in which I built an expert system for predicting pools coupons (1X2) using different Machine Learning techniques.

With the publication of this Data Set, we aim to explain in several tutorials some of the Machine Learning techniques that can be applied to predicting pools results, and also to encourage Machine Learning enthusiasts to dig in and share their own techniques and results for predicting football matches. Anyone who wants to collaborate and publish their experiments and results on this website, explaining the techniques and methods they used, can do so either through a post (always respecting the authorship of the experiments) or through the forum (https://jarroba.com/foro/)

Data Set

The Data Set with the football match results is provided as a "txt" file called "DataSetPartidos.txt", in which each line contains information about a football match played, separated by the characters "::". The meaning of each of these fields is as follows:

"idPartido::temporada::division::jornada::EquipoLocal::EquipoVisitante::golesLocal::golesVisitante::fecha::timestamp"

An example of the data for a football match would be the following:

4808::1977-78::1::8::Rayo Vallecano::Real Madrid::3::2::30/10/1977::247014000.0

The interpretation and meaning of this data is given below:

- idPartido (4808): Un identificador único de partido.
- temporada (1977-78): Temporada en la que se jugó el partido
- division (1): División en la que se jugo el partido (Primera '1', Segunda '2')
- jornada (8): Jornada en la que se jugó el partido
- EquipoLocal (Rayo Vallecano): Nombre del Equipo Local
- EquipoVisitante (Real Madrid): Nombre del Equipo Visitante
- golesLocal (3): Goles que marcó el equipo local
- golesVisitante (2): Goles que marcó el equipo visitante
- fecha (30/10/1977): Fecha en la que se jugó el partido
- timestamp (247014000.0): Timestamp de la fecha en la que se jugó el partido

Below we show the results of the football matches played on the first matchday of the 2014-15 season, in the first division:

33780::2014-15::1::1::Malaga::Athletic Club::1::0::23/08/2014::1408744800.0
33781::2014-15::1::1::Sevilla::Valencia::1::1::23/08/2014::1408744800.0
33782::2014-15::1::1::Granada::Deportivo::2::1::23/08/2014::1408744800.0
33783::2014-15::1::1::Almeria::Espanyol::1::1::23/08/2014::1408744800.0
33784::2014-15::1::1::Eibar::Real Sociedad::1::0::24/08/2014::1408831200.0
33785::2014-15::1::1::Celta de Vigo::Getafe::3::1::24/08/2014::1408831200.0
33786::2014-15::1::1::Barcelona::Elche::3::0::24/08/2014::1408831200.0
33787::2014-15::1::1::Levante::Villarreal::0::2::24/08/2014::1408831200.0
33788::2014-15::1::1::Real Madrid::Cordoba::2::0::25/08/2014::1408917600.0
33789::2014-15::1::1::Rayo Vallecano::Atletico de Madrid::0::0::25/08/2014::1408917600.0

We chose to publish this Data Set in this format because it is simple and similar to other data sets, such as the MovieLens and Netflix ones for recommender systems. With this format, it is very easy to parse regardless of the programming language you use to build your result prediction expert system.

For any collaboration, proposal, improvement, etc., please contact us at jarrobaweb@gmail.com or through the forum.