Chess tactics play a crucial role in chess games. Still, despite their importance, there is a gap in the existing literature regarding automatic methods for estimating the difficulty of tactical chess problems. In this master thesis, we aim to fill this gap by developing an approach that combines advanced artificial intelligence techniques, machine learning algorithms, and our chess knowledge to predict the difficulty of tactical chess problems with sufficient accuracy.
To achieve this, we used heuristic search algorithms to analyze the state space of the problem. Using the state-of-the-art open-source chess engine, we build a meaningful search tree that simulates a human approach to problem-solving. In addition to the meaningful tree, we extracted features by identifying a wide range of strategic and tactical chess motifs.
To train the model, we used a large problem set with established difficulty ratings from the chess platform Lichess. We analyzed the model's performance using different feature sets and gained a deeper insight into the difficulty factors in tactical chess problems. Our model has shown good enough accuracy to be used in practical applications, e.g. to help select suitably difficult problems for personalized chess training, or to analyze past games to estimate the difficulty of positions where a player has made a tactical error.
|
