Enhanced Decision Tree Algorithm for Data Streams using adaptation of Concept Drift

Construction of a decision tree is a well researched problem in data mining. Mining of streaming data is a very useful and necessary application. Algorithms such as VFDT and CVFDT are used for decision tree construction, but as a lot of new examples are added, a new optimal model needs to be constructed. Here in this paper, we have provided an algorithm for decision tree construction which uses discriminant analysis, to select the cut point used for splitting tests, thus optimizing time complexity from O(nlogn) to O(n). We have also analyzed several learning strategies such as dynamic ensemble, contextual, forgetting and detection approaches. We have also discussed handling of concept drift which occurs due to gradual change in the data set using the naive Bayes classifer at each of the inner node.