Classification of Chess Games: An Exploration of Classifiers for Anomaly Detection in Chess

Chess is a strategy board game with its inception dating back to the 15th century. The Covid-19 pandemic has led to a chess boom online with 95,853,038 chess games being played during January, 2021 on lichess.com. Along with the chess boom, instances of cheating have also become more rampant. Classifications have been used for anomaly detection in different fields and thus it is a natural idea to develop classifiers to detect cheating in chess. However, there are no specific examples of this, and it is difficult to obtain data where cheating has occurred. So, in this paper, we develop 4 machine learning classifiers, Linear Discriminant Analysis, Quadratic Discriminant Analysis, Multinomial Logistic Regression, and K-Nearest Neighbour classifiers to predict chess game results and explore predictors that produce the best accuracy performance. We use Confusion Matrix, K Fold Cross-Validation, and Leave-One-Out Cross-Validation methods to find the accuracy metrics. There are three phases of analysis. In phase I, we train classifiers using 1.94 million over the board game as training data and 20 thousand online games as testing data and obtain accuracy metrics. In phase II, we select a smaller pool of 212 games, select additional predictor variables from chess engine evaluation of the moves played in those games and check whether the inclusion of the variables improve performance. Finally, in phase III, we investigate for patterns in misclassified cases to define anomalies. From phase I, the models are not performing at a utilizable level of accuracy (44-63%). For all classifiers, it is no better than deciding the class with a coin toss. K-Nearest Neighbour with K = 7 was the best model. In phase II, adding the new predictors improved the performance of all the classifiers significantly across all validation methods. In fact, using only significant variables as predictors produced highly accurate classifiers. Finally, from phase III, we could not find any patterns or significant differences between the predictors for both correct classifications and misclassifications. In conclusion, machine learning classification is only one useful tool to spot instances that indicates anomalies. However, we cannot simply judge anomalous games using only this method