This research explores the relationship between visual attention and chess performance through an analysis of eye-tracking data collected from 30 participants engaging with 15 chess puzzles of varying difficulty levels. By constructing saliency maps and conducting correlation analyses, we identified that successful players tend to focus on critical areas of the chessboard. Furthermore, we have found 22 out of 35 general features, 11 out of 26 Important squares features, and 40 out of 55 Attention map features to be statistically significant. To predict puzzle-solving success, we evaluated three predictive models: Puzzle-agnostic (puzzle independent), Puzzle-informed (with additional features encoding puzzle information), and Puzzle-driven (individual feature set for each puzzle). The Puzzle-driven model, performing feature selection for each puzzle individually, achieved the highest accuracy of 0.8911 with a standard deviation of 0.0452. These findings not only highlight the importance of visual attention in chess performance but also showcase the value of the newly introduced complex features. Both the Important squares and the Attention maps can be viewed as extensions of ideas rooted in earlier literature, particularly chunk theory. Their success demonstrates the possibilities of applying chunk theory beyond theoretical explanations, enabling its practical use in real-world tasks such as feature construction and model development. • There is a notable link between player’s visual attention and puzzle- solving accuracy. • Visual attention can be used to predict the accuracy in Mate-in-One chess puzzles. • Prediction accuracy can reach 78% (Puzzle-agnostic model), 80% (Puzzle- informed model), and 89% (Puzzle-driven model).
Sukhorukov et al. (Sun,) studied this question.