Study On Algorithm Of Decision Tree Learning Based On Rough Set

Decision tree learning is one of the widly used and practical methods for inductive inference. Decision tree represents the information of original dataset by an explicit tree model. The non-predictive parts of a decision tree should be eliminated or pruned, but it is very difficult to estimate precisely when to stop growing the tree. It always depends on the expert knowledge or prior knowledge. This reduces the intelligent level of the learning process, and limits its application. Rough Set can analyze,process and acquire potential knowledge from imprecise, inconsistent and incomplete data. As a result, It can raise the automation degree of the knowledge acquisition process. Its validity has been conformed from the successful uses in various science and engineering domains in recent years. According to it, Rough set Theory is introduced in decision tree.The main researches and contributions are as follows:1. This paper has explained the basic concept of decision tree building at first. These typical algorithms are summarized and compared, and the estimation criterion of decision is presented.2. By analyzing the algorithm of decision tree based on rough set, Compared with the entropy-based ID3, the former method is simpler in the structure.Method of decision tree based on coordination degree is developed. Based on the basis of rough set coordination degree was used to select split-attributes. Condition certainty degree, which has the character of part optimization, could control the growing of decision tree. The process of tree building is analyzed by an example, and the experiments show that the method is feasible. The problem which the threshold of pruning is setted by expert is solved in the decison tree building.3. The problem of incremental knowledge acquisition is studied in this paper. The incremental decision-tree Pre-pruning algorithm is developed in this paper. The new idear is proved feasible by experiment. The computation complexity is analysed quantitatively. The incremental data in decision tree is realized, which can avoid the voluminous expense of decision tree rebuilding. It has better effect in little incremental data, and has discussed the next research direction.