Abstract
Chess games clustering poses the challenge of accurately grouping games with similar strategies and positions, especially when the openings are similar. Previous research has used Portable Game Notation (PGN) as a feature for clustering, but its emphasis on move order can limit position transposition. This research addresses this limitation by evaluating Forsyth-Edwards Notation (FEN), which focuses on board position, as an alternative. Hierarchical clustering with complete linkage and K-means clustering were used to analyze 100 chess games at move depths of 20, 30, 40, and 60. Both methods effectively cluster games involving the English Opening and the Queen's Gambit Declined, with FEN providing slightly better differentiation than PGN. However, challenges remain in grouping French Defence variations, especially the Poulsen Attack and variations with 6.a3, due to positional similarities. This study underlines the robustness of FEN for clustering tasks and its compatibility with hierarchical clustering, highlighting the important role of move depth. The results provide a basis for refining clustering methods and using larger data sets to deepen insights into chess strategies.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call