Recognition of Recurrent Movement Patterns of Football Players via Machine Learning

Team sports like football require a specific training of athletes’ movement patterns. Thus, recognising such patterns for each role can be of particular interest for trainers, in order to correctly plan tactical strategies and training activities. A methodology to discover the most frequent velocity and acceleration patterns of football players is here proposed. It exploits data collected via wearable GPS and IMU sensors during competitive professional matches and machine learning analysis techniques. Movements are first clustered and labeled and, then, considered in fixed length sequences. These are, in turn, compared among each other and grouped using a hierarchical clustering algorithm. Lastly, Longest Common Subsequence algorithm is used to find the common patterns of players’ movements. A total of 35 frequent patterns have been detected for the five main football players’ roles striker, central midfielder, side back, wing forward and central back.