Analysis of the frequency of football players’ movement patterns by means of Lorentzian Self-Organizing Maps
Author(s)
Melloni, DANIELE
Date Issued
2025
Type
article
Volume
159
Start Page
111703
Abstract
The recognition of frequent movement patterns of football players is of particular interest to assess, from a tactical perspective, the performance of both single individuals and the team as a whole. This paper introduces a novel variant of Self-Organizing Maps (SOMs), called Lorentzian Self-Organizing Maps (LSOMs). Whereas current research in sports science addresses a variety of relevant tasks, the dynamic and evolving nature of player movements patterns throughout a match remains underexplored. Starting from players’ velocity and acceleration data, acquired through a sensorized wearable vest during competitive matches, movement units are built, by clustering ranges of similar velocity-acceleration pairs. Subsequently, this units are grouped to form patterns of varying length, and the most frequent ones for five different players’ role are detected and extracted. Its comparison against other state-of-the-art clustering algorithms shown a consistent improvement. The testing has been performed by relying on 3 performance indexes, namely normalized mutual information, clustering accuracy and adjusted rand index. LSOMs resulted in the highest average scores on all the three metrics. Furthermore, in order to assess the tactical relevance of the proposed approach, a questionnaire was administered to 5 professional football trainers, about the goodness of the clustering proposed by LSOMs, against the other clustering techniques. The trainers feedback
suggested that LSOMs can be profitably used to analyze players’ frequent movement patterns and that this kind of analysis can be helpful in finding non-trivial connections between the roles and the movements and, consequently, in gaining additional tactical insights.