Evaluation of match results of five successful football clubs with ensemble learning algorithms

Abstract

ABSTARCT Purpose: Football, one of the most popular and loved sports branches, always keeps its excitement, ambition, passion, joy and sadness together. European football, the football capital, is an attraction for fans and footballers. In this study, the official match results (league, country cup, European cup) of five successful football clubs (Bayern Munchen, Barcelona, Juventus, Manchester City, Paris Saint Germain) in the five major leagues of European football (La Liga, Premier League, Serie A, Bundesliga, Ligue 1) were evaluated. Method: For this analysis, ensemble learning algorithms (AdaBoost, Bagging) and machine learning algorithms (Naive Bayes, artificial neural networks, K-nearest neighbor, C4.5/Random forest/Reptree decision tree) were used. In addition, the attributes that play an active role in the classification of the match results of five successful football clubs were determined with the Symmetrical Uncertainty feature selection algorithm. Results: As effective attributes, "Conceded goal," "Half time result," "Scoring first" and "Shooting accuracy" attributes revealed to be common for five successful football clubs. In general, it was observed that ensemble learning algorithms gave successful results and AdaBoost/ANN algorithm was determined as the most successful. On the basis of football clubs, the most successful classification result was achieved for Barcelona with a rate of 99.3%. Conclusions: Obtained outputs from Ensemble learning and feature selection help sport researchers and football club planners understand and revise the match results of current football match strategies. The study has mainly twofold: to find best performer ensemble and machine learning algorithm(s) for classifying match results and to extract important features on match results.