The Research On Data Stream Classification Algorithm Based On Knowledge Integration

Not only the result of classification will be changed for the generated concept drift ,but also the model of knowledge in data stream. In this paper,an algorithm(Knowledge Integration Dynamic Decision Tree ,KIDDT) based decision tree knowledge Integration is proposed to deal with the problem. The data block will be captured properly to construct the Incomplete model of decision tree, and then integrated these parts of the sub-knowledge model. The weight thought in the algorithm of ensemble will be introduced into KIDDT for the attribute decision making. The integrated knowledge model namely the finally decision tree which includes relatively complete and unified knowledge structure.It sums up multiple classifiers with Incomplete knowledge model. KIDDT just using the final integrated decision tree with no shadow node when it predicting samples.KIDDT system is developed in the open source data mining software WEKA and JAVA environment Myeclipse6.5. Finally,the superiority and effectiveness of KIDDT algorithm is verified by a series experiments with stream dataset.