An Improved Algorithm of Decision Trees for Streaming Data Based on VFDT

Abstract

Decision tree is a good model of Classification. Recently, there has been much interest in mining streaming data. Because streaming data is large and no limited, it is unpractical that passing the entire data over more than one time. A one pass online algorithm is necessary. One of the most successful algorithms for mining data streams is VFDT(Very Fast Decision Tree).we extend the VFDT system to EVFDT(Efficient-VFDT) in two directions: (1)We present Uneven Interval Numerical Pruning (shortly UINP) approach for efficiently processing numerical attributes. (2)We use naive Bayes classifiers associated with the node to process the samples to detect the outlying samples and reduce the scale of the trees. From the experimental comparison, the two techniques significantly improve the efficiency and the accuracy of decision tree construction on streaming data.