earticle

영어

In this paper, we propose the clustering-based ensemble management system which extracts new data patterns from the input streaming data by using clustering and generates new classification models only when a certain amount of data has been collected in a cluster. The clustering-based ensemble management system can reduce the number of the data labeling and keep the accuracy of the existing ensemble. The clustering-based ensemble management system collects similar patterned data from the input streaming data for building a cluster. The clustering-based ensemble management system performs the data labeling for the each cluster only when a certain amount of data has been collected in the cluster. The data labeling by experts goes on to generate the new classification model to be added to the ensemble. The clustering-based ensemble management system applies the K-NN technique to classification model units in order to keep the accuracy of the existing system while it uses a small amount of data. The efficiency of the clustering-based ensemble management system proposed in this paper is shown by the simulated results for benchmarks comparing with the existing ensemble techniques.