Dual weighted extreme learning machine for imbalanced data stream classification

Data streams with class imbalance occur usually in many real applications. Online sequential learning is one of the effective methods for classifying data stream with class imbalance. This paper proposes a dual-weighted online sequential extreme learning machine (dw-ELM) method to solve it. On the basis of online sequential extreme learning machine, the proposed dw-ELM method analyzes the distribution characteristic of data in view of time and space, and gives an adaptive dual weighting scheme to tune the weights at both the time level and the space level. Extensive experimental evaluations on 10 imbalanced datasets indicate that the proposed dw-ELM method outperforms several comparing methods in terms of G-mean and F-measure metrics. Moreover, the proposed dw-ELM method remains superior classification performance in the presence of highly dynamic class imbalance.