Chess Moves Prediction using Deep Learning Neural Networks
Hitanshu Panchal, Siddhant Mishra, Varsha Shrivastava
Abstract
Chess is a game that is popular for high intelligence and strategic thinking. There has been a lot of research on chess for predicting chess moves, applying chess game theory, and automating chess games. The art of playing chess using computer vision can be implemented using various learning algorithms. A class of Deep Learning has the ability to solve problems of predicting chess moves although facing the necessity of huge datasets. The traditional chess algorithm Minimax with the Convolutional Neural Network can perceive and learn the patterns and rules in chess i.e., identification of some small and native tactics of the game, and should be trained on this method with appropriate functions for smarter universal play. CNN when trained with appropriate architecture and validation data can learn to function based on the reasoning in complex logical tasks. Training on 15,00,000 board states in the dataset which is a board state represented as 8x8x14 dimensions. Each board state is given as an input to the input layer of the Convolutional Neural Network. The CNN model tested and validated against the stockfish chess engine achieved the best accuracy of 39.16% for board evaluation. However, this doesn’t reflect the actual accuracy of the model since the evaluation by the model is relative for two different board states. CNN learning the game of chess and based on the result of chess is essentially pre-computation on a given situation.