An Ensemble Approach for Handling the Novel Instances in Dynamic data

Many algorithms have already been designed to control concept drift problem in dynamic data. In this paper, authors propose the “Weighted Novel Instances Detection”. This approach handles the novel instances with diversified ensemble learning. It uses various concepts such as diversified online ensembles, instance weighting and detection of novel instances. The final prediction of an instance is the class predicted by an ensemble. Experimental results prove that it efficiently classifies recent examples.