Modeling Chess Strategy by Classifier Based on Imbalance Learning and Application in Computer Chinese Chess

Computer chess game (CCG) is an important topic in the field of artificial intelligence. This technique is widely used in some entertainment PC games and chess games on different platforms. Most CCG systems are developed based on the combination of game tree searching and evaluation functions. When using game tree searching method, the level of the computer player depends on the searching depth. However, deep game tree searching is time-consuming when the games are applied on some mobile platforms such as mobile phone and PDA. In this paper, a novel method is proposed which models Chinese chess strategy by training a classifier. When playing chess games, the trained classifier is used to predict good successor positions for computer player. The training procedure is based on imbalance learning and it uses Chinese chess game records as the training sets. Specifically, the training sets extracted from game records are imbalanced; therefore, imbalance learning methods are employed to modify the original training sets. Compared with the classical CCG system, this new method is as fast as 1-level game tree search when playing games, and it contains an offline learning process. Experimental results demonstrate that the proposed method is able to model Chinese chess strategies and the imbalance learning plays an important role in the modeling process.