An Algorithm Of Discretization Based On Entropy In Application Of Decision Tree

In all the methods of classification, the decision-tree method is used vastly. But by the reason of its disadvantages in dealing with continuous datas, it does not has a good ability of operating. To resolve this, a new algorithm has been advanced by the thinking of game. The discretization is considered to be a game model, and by its own way, it can elicit the result that we need.In this thesis, it has been proved that the nash poise in this game model is the result we need. In forth chapter, the algorithm that how to get the nash poise is advanced. Firstly, it can choose a better way to find in the front, secondly, it can adjust the direction dynamicly by the information gained during finding the poise, thirdly, it can reduce the domain that contain the nash poise. So it can reduce time that finding the poise.In order to reduce the load of computation, it has been improved by some measures which has been proved effective. Because of the not known of existence in all game-model, it has happen that the computation would be get into the difficulties of limitless circulate. In order to avoid this, it should define a halt rule based on the evolvement, and evaluate the result of the discretization. Then we can get a satisfied result, and take a good basis data of classifying.The experimentation in this thesis has two part. One is the contrast between the classic way of finding nash poise and the one the paper advanced. The other is the contrast between the EBD algorithm and the one the paper advanced. Then we can improve this algorithm to reduce complexity of the computing by the fact contition. This new algorithm not only maintains simplicity, consistency and accuracy but also is easily operated.