Changing Lineup Classifier Ensemble for Drifting Imbalanced Data Streams

Currently, most data requires stream processing. This kind of processing causes new challenges in classifier learning method, such as data difficulties for example data imbalance since we cannot observe the entire set of data simultaneously or concept drift, which is the possibility of changes in the probabilistic characteristics of the processed data. These challenges were addressed by proposing a new learning algorithm that maintains consistency with the current data distribution by training a classifier ensemble with the changing lineup and applying dedicated mechanisms to account for the imbalance of the incoming data stream. The proposed IMB-WAE (Imbalance Weighted Aging Ensemble) method has been evaluated on the basis of exhaustive computer experiments conducted on a large number of real and synthetic data streams assessing the impact of its parameters on classification quality and confirming its quality compared to the state-of-the-art algorithm.