Research On Constructing Decision Tree Based On Rough Degree Measurement Of Knowledge

Classification is an important research branch of data Mining.Decision tree is one of the widely used approachs in classification,which being widely researched and applied.However,decision tree has some disadvantages shch as variety bias,poor anti-noise capability etc,and optimization of decisiong tree has become one of the research hotspots. The dissertation focuses on constructing approachs and optimization of decision tree based on rough set theory, and the main achievements are as follow:(1)An overview and analysis of classical and optimized decision tree constructing approachs is put forward.(2) The decision tree algorithm KRD is proposed, which uses rough degree of knowledge as the criterion of selecting attributes. The scale and classifying precision of the constructed decision tree in this method has been improved comparing to the ID3 algorithm's.(3) The hybrid decision tree algorithm KRDH is presented,which confirms the best attribute according to the varies of the rough degree of knowledge,because of the noise influence and the problem of feebleness relativity in the reality data.(4) In order to improve the anti-noise capability of decision tree, KRDlc algorithm is presented,which is based on statistical model.It makes use of rough degree of knowledge to select splitting attributes and according to the leaf control parameter to prun the decision tree,so the attribute selection will be less influenced by noise and overfitting can be avoided and the scale of decision tree can be decreased.