Abstract: In game development, if scene design as the focus is designed one by one, a lot of time and capital will be spent, and players stick low degree. Therefore, a game level automatic generation method based on data mining and data fusion technology is presented. Firstly, the method preprocesses the game play data by using the discretization of Boolean logic combined with rough set theory, puts forward an attribute reduction algorithm based on information gain to eliminate the redundant attributes. Secondly, a decision tree is constructed by using the ID3 algorithm to build an evaluation model of game difficulties. Then the D-S data fusion algorithm is used to get the data reflecting the players behavior, and the data is processed with the decision tree to get the difficulty level of the game for players. Finally, the game level parameters are modified to automatically generate the game level according to the difficulty level. By the experiment of Sokoban game for example, the results show that the method is effective to reduce development cost and to improve playability.

Key words: data mining, data fusion, attribute reduction, game level, automatic generating