Detecting switching leadership in collective motion

Area of Science:

• Complex Systems Science
• Information Theory
• Collective Behavior Analysis

Background:

• Detecting causal relationships in complex systems from time series data is challenging.
• Understanding evolving causal links, like leadership in groups, requires advanced methodologies.

Purpose of the Study:

• To develop a robust method for detecting leadership switches in collective behavior.
• To augment transfer entropy with a fitness function for time series partitioning.

Main Methods:

• Augmented transfer entropy with a novel fitness function.
• The fitness function optimizes time series partitioning by measuring information outflow.
• The method rewards large sample sizes and normalizes information.

Main Results:

• Successfully detected leadership switches in collective behavior with over 90% accuracy.
• Effective detection achieved even with short time durations in information-rich interactions.
• Identified leadership instances correlated with ball possession in a soccer dataset.

Conclusions:

• The proposed method robustly detects leadership switches in collective behavior.
• This approach offers a powerful tool for analyzing dynamic causal relationships in complex systems.
• Findings have implications for understanding group dynamics in various fields.