This chapter introduces machine learning-based analysis of multi-agent trajectoriesMulti-agent trajectories in basketball. It addresses the challenge of comprehending multi-agent behaviors, which are typical in team sports, and characterized by intricate interactions and cognitive processes. The chapter outlines how machine learning methods can effectively analyze such behaviors, despite the inherent difficulty in modeling the behaviors and interpreting these non-linear models. We highlight two primary strategies: learning-based feature and rule extraction, and generating and controlling behaviors from models. The first involves visualizing representations and identifying underlying structures, while the second allows for the simulation and control of potential and hypothetical scenarios. The chapter concludes by discussing the practical implications of these methods, emphasizing their utility in enhancing our understanding of multi-agent behaviors in basketball.