Modeling for Classifying Data Streams with Concept Drift
Joung Woo Ryu, Jin-Hee Song · Advanced science and technology letters · 2014
We propose a novel methodology of maintaining a classification model on streaming data, such as sensor data or log data. Our approach uses an ensemble for classifying data streams which consists of a set of classifiers. New classifiers of the ensemble on streaming data will be generated dynamically according to the estimated distribution of streaming data instead of periodically building them. Also our approach is able to deal with changes of class distribution as well as changes of data distribution. We compared the results of our approach and of the previous approach which can only deal with changes of data distribution. In experiments with 10 benchmark data sets, our approach produced an average of 3.61% higher classification accuracy.