Stochastic configuration networks for imbalanced data classification

Stochastic configuration networks (SCNs), as a class of advanced randomized learner models, play an important role in predictive data analytics. Given an imbalanced data classification task, the original SCN classifiers may fail to provide satisfied performance because of the density difference of data distribution. This paper contributes to a development of imbalanced learning for SCNs (IL-SCNs) classifier design with skewed class distribution. Concretely, a balancer is proposed and used in IL-SCNs to compromise between the majority class and the minority class. In addition, a fast computation algorithm is adopted to update the output weights, which achieves lower computation complexity of IL-SCNs. Experimental results show that IL-SCNs significantly outperforms the existing state-of-the-art learning models.