[Paper Review] Sensor Analytics in Basketball

Rodolfo Metulini, Marica Manisera+1 more|arXiv (Cornell University)|Jul 4, 2017

Sports Analytics and Performance3 citations

TL;DR

This paper applies sensor analytics to basketball using GPS-tracked player trajectories, employing convex hulls and k-means clustering to identify movement patterns linked to team performance. It finds that offensive plays feature significantly larger player spacings—measured by convex hull area—than defensive plays, indicating a strategic spread to enhance scoring opportunities.

ABSTRACT

A new approach in team sports analysis consists in studying positioning and movements of players during the game in relation to team performance. State of the art tracking systems produce spatio-temporal traces of players that have facilitated a variety of research aimed to extract insights from trajectories. Several methods borrowed from machine learning, network and complex systems, geographic information system, computer vision and statistics have been proposed. After having reviewed the state of the art in those niches of literature aiming to extract useful information to analysts and experts in terms of relation between players' trajectories and team performance, this paper presents preliminary results from analysing trajectories data and sheds light on potential future research in this field of study. In particular, using convex hulls, we find interesting regularities in players' movement patterns.

Motivation & Objective

• To identify regularities in player movement patterns using sensor-derived trajectories in basketball.
• To link spatial positioning and movement dynamics to team performance outcomes.
• To develop visual and analytical tools that support coaches and analysts in interpreting tactical behavior.
• To explore how player spacing, measured via convex hulls, differs between offensive and defensive phases.
• To lay the foundation for future data-driven analysis of team sports using multivariate trajectory data.

Proposed method

• Utilizes GPS-tracked spatio-temporal trajectories of players from real basketball games.
• Applies k-means clustering to group time instants based on inter-player distances, with k=8 selected via between-to-total deviance ratio.
• Employs multidimensional scaling (MDS) to visualize player positions in 2D space while preserving average inter-player distances.
• Defines game phases as homogeneous moments in terms of player spacing and team positioning.
• Uses convex hull area as a geometric measure of player spread across the court.
• Analyzes the time series of convex hull areas to compare offensive and defensive strategies between teams.

Experimental results

Research questions

• RQ1 How do player positioning patterns differ between offensive and defensive phases in basketball?
• RQ2 To what extent does player spacing, measured by convex hull area, correlate with team performance?
• RQ3 Can clustering of trajectory data reveal distinct tactical formations or movement patterns?
• RQ4 Do defensive teams successfully reduce the convex hull area of offensive teams during play?
• RQ5 What regularities in movement patterns can be extracted from sensor data to inform coaching decisions?

Key findings

• Cluster 1 (C1), Cluster 2 (C2), Cluster 3 (C3), and Cluster 4 (C4) predominantly represent offensive plays, with 85.88%, 85.91%, 73.93%, and 84.62% of their time instants classified as offensive, respectively.
• Cluster 6 (C6) is strongly associated with defensive actions, with 85.07% of its time instants classified as defensive.
• Offensive phases exhibit significantly larger convex hull areas than defensive phases, indicating greater player spread on the court.
• The average convex hull area is larger during offensive plays, suggesting a strategic effort to create space for scoring opportunities.
• Multidimensional scaling (MDS) visualizations confirm substantial differences in player positioning structures across clusters.
• The study identifies a consistent pattern where offensive teams maintain greater spatial dispersion compared to defensive teams, who aim to reduce such spacing.