A tactic-based prediction system for football matches

Organizations of sports competitions are among the important activities that people spend their time on. Among the sports competitions, football has the highest popularity rate in the world. Many football competitions are organized both on a regional basis and on an international basis. The results of football matches are very important for the fans of the football team, and they can be happy or sad according to the results of the team they support in their daily lives. As a result of developments in artificial intelligence and machine learning, predictions made in many fields include the prediction of football results. Considering the football betting system, which also has economic returns, the prediction of football match results has become important. This study presents a tactical-based forecasting approach to be presented for the first time in the literature to our knowledge, to predict results in Italy-Serie A, England-Premier League and Germany-Bundesliga leagues. In this approach, players' on-field placements and market values are automatically obtained from web pages and made available for data mining methods. The data mining phase compares different methods such as Bagging, Iterative Classifier Optimizer and Deep Learning, and it is seen that the three methods give similar results in different leagues. An accuracy of about 60% is achieved in data mining methods. When the features used in machine learning methods are examined for all three leagues, it is the most important feature in predicting the team ranking. While Center-Back and Central Midfielder features stand out in Serie A, different features such as Right Winger and Goalkeeper are seen to be important in Premier League. In the Bundesliga, the Centre-Back and Centre-Forward features have had a greater impact on the matches.