Rough Set Theory In The Decision Tree

Classification is an important data-mining problem which has been studied extensively by machine learning community as a possible solution to knowledge extraction problem. One of the important classification techniques is decision trees. Why decision trees? Compared to a neural network or a Bayesian classifier, a decision tree is easily interpreted /comprehended by humans;Training a neural network can take large amount of time and thousands of iterations;Inducing decision trees is efficient and is thus suitable for large training sets;Decision tree generation algorithms do not require additional information besides that already contained in the training data;Decision trees display good accuracy as compared to other techniques. But decision trees also have some limitation. On the one hand, it can not delete irrelevant attributes;One the other hand, most decision trees can test only one variate on each node.In order to overcome these limitations, we import Rough Set techniques. Rough Set is one of the new logic and research ways. It is a new mathematical tool to deal with fuzzy and uncertain knowledge. It has strong knowledge obtaining ability. The main point of rough set is that it can integrate knowledge with classification and that knowledge is the ability of classifying the objects. Although it is effective in dealing with the imperfection knowledge, it is not omnipotent. It is weak in tolerance and generality, that is to say, it needs to integrate with other technology.Considering the advantage and disadvantage of the decision trees and RS, how about to combined the decision trees and RS? In this paper, a new alternation algorithm is offered, which restricts each of node containing the number of attributes, and then, choice attributes combination according to the redefined attribute dependability and the distance function based on condition entropy.Value reduction in rough set theory and decision tree in data mining are effectively used in the classification, which can be combined to generate a new minimal method based on value core to pollard the decision tree. Moreover, the standards of rule reduction were proposed to decrease the quantity of the reduced rules, and then, the classing rules weredealed with maximal method to ensure the consistency of the knowledge.