A Bayesian Networks Model for Football Matches Result based on Player Performance

Nazim Razali, Aida Mustapha and Muhaini Othman

Bayesian Networks (BN) is extremely useful in modeling probability distribution for predicting, reasoning, and decision making under uncertainty where the knowledge is represented in the form of graphical models called the Directed Acyclic Graphs (DAG). In sports analytics, BN has been primarily used to model team performance in predicting football matches result based on the presence the presence of individual players in a specific match. In this paper, we use BN to predict football matches results based on individual players performance as opposed to team performance. Apart from the individual player performance data, the dataset includes individual player rating, absence or present of players in a match, venue team performance and opposition performance. The dataset used was the Arsenal Football Club in English Premier League (EPL) for season 2014-2015 and 2015-2016 with leave-one-out validation technique. The results showed that the proposed player performance model is better than existing team performance, hence is recommended as new feature set for future football prediction model.

Volume 11 | 05-Special Issue

Pages: 827-833