An Ensemble-Based Informational Fast Resampling Method for Multiclass Imbalanced Data Stream Classification

Data stream is a common and infinite form of data in the era of big data. Ensemble learning can adapt to the dynamic update characteristics of data stream, thus having significant advantages in data stream classification. In addition, the most commonly used method for class imbalance is resampling, which can change the underlying distribution of data and reduce the imbalance ratio. Therefore, this paper proposes an ensemble-based Informational Fast Resampling Method (IFRM) for multiclass imbalanced data stream classification. Firstly, the issue of multiclass imbalanced data stream is analyzed. Secondly, the framework of the proposed method is introduced, which includes four parts: class imbalance judgment, samples selection, samples synthesis, and weighted ensemble. Finally, the performance of the proposed method is verified through various experiments. Theoretical analysis has shown that this method has low time complexity compared to other multiclass data stream classification algorithms. Furthermore, experimental results confirm the effectiveness of the informational fast resampling technique proposed in this paper.