Pairwise Learning to Rank for Chess Puzzle Difficulty Prediction

Machine learning has shown great success in various aspects of chess, particularly in game-playing engines such as AlphaZero. However, predicting the difficulty of chess puzzles remains a relatively unexplored area. In the IEEE BigData 2024 Cup: Predicting Chess Puzzle Difficulty competition, participants are asked to build a machine learning approach to predict the difficulty of chess puzzles. We present an approach that leverages deep learning and pairwise learning-to-rank techniques to estimate the difficulty of chess puzzles. Our method applies pairwise learning to rank approaches to simulate games between puzzles and uses the outcomes to estimate their Glicko-2 ratings. This approach achieved 4th place in the competition, demonstrating its effectiveness.