A Framework for Spatio-Temporal Graph Analytics In Field Sports

TL;DR

This work addresses the need to understand team dynamics in field sports using spatio-temporal GPS data rather than solely relying on individual performance metrics. It introduces a four-step framework that transforms tracking data into Time-Window Spatial Activity Graphs ($TWG$) by overlaying a spatial grid, segmenting data into rolling time windows, and applying graph analytics. The authors formalize TWGs, demonstrate betweenness centrality and Louvain-based community detection on Gaelic Football data, and extract insights about areas of activity and inter-area connectivity over time. The approach enables data-driven coaching and strategic planning by revealing how teams navigate space and adjust tactics throughout a match.

Abstract

The global sports analytics industry has a market value of USD 3.78 billion in 2023. The increase of wearables such as GPS sensors has provided analysts with large fine-grained datasets detailing player performance. Traditional analysis of this data focuses on individual athletes with measures of internal and external loading such as distance covered in speed zones or rate of perceived exertion. However these metrics do not provide enough information to understand team dynamics within field sports. The spatio-temporal nature of match play necessitates an investment in date-engineering to adequately transform the data into a suitable format to extract features such as areas of activity. In this paper we present an approach to construct Time-Window Spatial Activity Graphs (TWGs) for field sports. Using GPS data obtained from Gaelic Football matches we demonstrate how our approach can be utilised to extract spatio-temporal features from GPS sensor data

Figures (6)

• Figure 1: Overview of framework for spatio-temporal graph construction
• Figure 2: Sample of 25 nodes of the TWG for time windows: [0,5) (left) and [1,6) (right) in a selected game before edges aggregation.
• Figure 3: Spatial grid mapping of the pitch. Left: grid points returned by the library. Right: the result of the association of players' coordinates to their closest grid point.
• Figure 4: Betweenees centrality for 5 minutes rolling windows: [0, 5) (top-left), [1, 6) (top-right), [2, 7) (bottom-left), and [3, 8) (bottom-right).
• Figure 5: Community detection for 5 minutes rolling windows: [0, 5) (top-left), [1, 6) (top-right), [2, 7) (bottom-left), and [3, 8) (bottom-right). Black points represent coordinates not covered by players in the analyzed time window.