Single Agent Behavior Prediction in Soccer Using Linear Temporal Logic
Loading...
Authors
Arastehfar, Sana
Date
Type
thesis
Language
eng
Keyword
Sport Analysis , Soccer Analysis , Linear Temporal Logic , Behavior Prediction , Baysian Inference
Alternative Title
Abstract
Analyzing and comprehending the behavior of soccer players is vital for refining match analysis and optimizing team performance. The interpretability of such behavioral insights has significant potential to enrich decision-making processes. In this thesis, we introduce a novel framework to extract the long-standing behavior of a soccer player, using Linear Temporal Logic (LTL); a logic system dedicated to describing how discrete properties change over time. Using LTL as a foundational mathematical tool, we establish a means to describe time-evolving events as logical propositions that encapsulate temporal dynamics. Our approach involves using a sequence of positional data containing coordinates of the players and the ball and match data. We characterize the temporal dynamics of players by employing defined and interpretable features such as player location, opponent pressure, reachable teammates, \textit{etc.} Then we use these temporal attributes to subsequently translate them into logical propositions, which are used as inputs for a sophisticated Bayesian inference engine to compute LTL expressions. This work marks a stride in harnessing LTL to extract intelligible logical propositions by first deriving meaningful feature vectors which can be used by Bayesian methods to compute LTL formulas that illuminate single-player behavior, for example, one might figure out from these formulas that when a player is playing in the midfielder position, opponent pressure will be high for that player. We conduct a quantitative and qualitative analysis using the SkillCorner dataset for the 2022/2023 English Premier League to showcase the efficacy and expressive capacity of our methodology. Furthermore, we outline avenues for future research that can further expand upon and refine the proposed framework.