Machine learning-based identification of the strongest predictive features of scoring penalty kick in football

Abstract

In football, the penalty is the situation that has one of the highest chances of scoring a goal. However, the success of a penalty kick highly depends on many kinds of attributes, including the penalty-takers’ abilities, the amount of fan pressure, the minute of the match, and the current score. In this paper, 16 features extracted from penalty kick positions, penalty-takers’ information, and match-day preferences and machine learning used to predict penalty kick result. Moreover, we revealed the most important feature combination that significantly affected the success of a penalty kick. The proposed method was trained with 120 and tested with 50 penalty kicks from the Turkish Super League in terms of classification accuracy and polygon area metric. We concluded that the result of a penalty kick can be predicted with a mean classification accuracy and mean polygon area metric rates of 79.80% and 0.60 using the k-nearest neighbor classifier.