Efficient Gaussian Decision Tree method for concept drift data stream

Decision tree has become one of the most accepted tools for mining data streams after Hoeffding tree was anticipated in the literature. The most vital point of constructing the decision tree is to find out the best attribute to split the considered node. Numerous methods to resolve this problem were presented so far, however, there are some shortcomings such that they are either mathematically not justified or time-consuming. In this paper, a new decision tree algorithm named Adaptive Gaussian Decision Tree (AGDT) is proposed that employs a statistical method (Gaussian Bound) for determining the best attribute in a node. This statistical method ensures that the finest attribute chosen in the considered node using a limited data sample is the same as it would be in the case of the complete data stream hence solving the storage constraints associated with the data stream. In order to handle the concept drift problems AGDT uses fixed-size window to determine which nodes are aging and may need updating. During implementation it is observed that the proposed decision tree method provides a greater classification accuracy, compared with the existing algorithm with the same probability for concept drift data streams.