Learning decision trees from time-changing uncertain data streams

Abstract

In this study, we study the problem of classifying uncertain data streams. Based on CVFDT algorithm, we proposed a novel algorithm, namely uCVFDTc, to learn very fast decision trees from uncertain data streams with concept drift. In training phase, the uCVFDTc algorithm uses Hoeffding bound theory to yield fast and reasonable decision trees. In classification phase, at tree leaves it uses Uncertain Naive Bayes (UNB) classifiers to improve classification performance. Experimental results showed that uCVFDTc had strong ability to learn from uncertain data streams and cope with concept drift; the use of UNB at tree leaves had improved the performance of uCVFDTc, especially the ability to handle concept drift. © 2013 Asian Network for Scientific Information.