A Novel Ensemble Classification for Data Streams with Class Imbalance and Concept Drift

The processing of streaming data implies new requirements concerning restrictive processing time, limited amount of memory and one scan of incoming instances. One of the biggest challenges facing data stream learning is to deal with concept drift, i.e., the underlying distribution of the data may be evolving over time. Most of the approaches in the literature are under the hypothesis that the distribution of classes is balance. Unfortunately, the class imbalance issue is common in the real-world. And the imbalance issue further increases the difficulty of solving the concept drift problem. Motivated by this challenge, a novel ensemble classification for mining imbalanced streaming data is proposed to overcome both issues simultaneously. The algorithm utilizes the under-sampling and over-sampling techniques to balance the positive and negative instances. Moreover, dynamic weighting strategy was adopted to deal with concept drift. The experimental results on synthetic and real datasets demonstrate that our proposed method performs better than competitive algorithms, especially in situations where there exist concept drift and class imbalance.