Inverse reinforcement learning to evaluate goal impact in ice hockey
2025 (English)Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE credits
Student thesisAlternative title
Invers reinforcement learning för utvärdering av ishockey : Inverse reinforcement learning in ice hockey (Swedish)
Abstract [en]
This study explores the application of machine learning through reinforcement learning techniques to assess individual player performance in ice hockey based on outcomes in the National Hockey League (NHL). The primary objective was to develop a value function trained using rewards linked to specific game outcomes, enabling a detailed analysis of player impact on the game. The methodology involved collecting game data, defining rewards based on match outcomes, and training a value function to understand and predict how individual player actions influence the possibility of winning. The results demonstrate that the trained value function effectively estimates player contributions and offers a nuanced view of their impact on game dynamics. This research confirms the potential of advanced analytical techniques in sports analytics, providing a foundation for future studies to refine player evaluation metrics and improve team strategy based on quantitative data.
Place, publisher, year, edition, pages
2025. , p. 41
Keywords [en]
reinforcement learning, machine learning, hockey analytics
National Category
Computer Vision and Learning Systems
Identifiers
URN: urn:nbn:se:liu:diva-215617ISRN: LIU-IDA/LITH-EX-A--25/006--SEOAI: oai:DiVA.org:liu-215617DiVA, id: diva2:1977040
Subject / course
Computer Engineering
Presentation
2025-02-28, Charles Babbage, B-huset Linköpings universitet, Linköping, 15:20 (English)
Supervisors
Examiners
2025-06-272025-06-252025-06-27Bibliographically approved