Uncertain Data Stream Classification with Concept Drift

Abstract

In big data era, the data on the Internet is growing at an exponential rate. The uncertainty of data due to privacy protection, data loss, network errors and so on is very common. In data stream system, data arrive at continuously and can't be obtained all. In addition, the concept drift occurs often in the data stream. So we need construct an incremental classification model to deal with uncertain data stream classification with concept drift. This paper presented Weighted Bayes based Very Fast Decision Tree for Uncertain data stream with Concept drift-WBVFDTUC algorithm. The algorithm can analyze uncertain information quickly and effectively in both the learning stage and classification stage. In the learning stage, it uses Hoeffding bound theory quickly construct a decision tree model for uncertain data stream. In the classification stage, it uses the weighted Bayes classifier in the tree leaves to improve the performance of the classification. The use of sliding window and replacing tree ensure the algorithm can deal with concept drift phenomenon. Experimental results show that the proposed algorithm can very quickly learn uncertain data stream and improve the classification performance of the model.