Alex Anuszkiewicz

Chess Outcome Prediction

One of four outstanding projects endorsed by the professor for CS365 Spring '24. Directly applied supervised machine learning methods to predict the ultimate outcome of chess games. A least-squares regression was utilized to show correlation between variables and identify important predictors. Constructed a logistic regression model that could classify chess game outcomes with 84% accuracy given the ELO difference, time limit of the game, the opening played, and the final state of the chess board. Built a convolutional neural network (CNN) model that predicted game outcomes given an arbitrary board state, with decreasing accuracy relative to the number steps away from a final board state. Achieved an accuracy of 85.59% for endgame scenarios, which declines to a 59.33% average accuracy. This project underscores the inherent complexity of chess, highlighting predictive challenges due to its combinatorial nature and lengthy strategic development. A decent CNN model has been developed that suggests areas for further predictive model enhancement and method evolution.

Github Link