Using (copula) regression and machine learning to model and predict football results in major European leagues

Abstract

In this manuscript, we compare classical univariate regression approaches with copula models explicitly accounting for the dependency structure as well as with modern machine learning techniques in the context of modelling and predicting of football results in the major European leagues. Particularly, we want to present an extensive data set compiled from publicly available sources containing data and match results from the first men's football divisions from England, France, Germany, Italy, Spain (often referred to as the ``big five''), the Netherlands and Turkey. We introduce several modelling approaches to predict upcoming matches and compare their predictive strengths. The gather data set is presented in detail and made publicly available to motivate further work and modelling ideas.