Ensemble learning-based online sequential pre-interference extreme learning for concept drifting and class imbalanced data streams

Abstract

With the rapid development of data-driven technologies, real-time data streams not only exhibit concept drift but are also frequently accompanied by class imbalance problems. To address these challenges, this paper proposes an online sequential pre-interference layer extreme learning machine (OS-PIELM). The proposed model introduces a pre-interference layer between the input layer and hidden layer of the original OS-ELM to enhance nonlinear feature representation through kernel-like transformation of sequential data, thereby improving the discriminative ability of different classes. Furthermore, an adaptive forgetting factor and a Gmean-based concept drift detection mechanism are incorporated into OS-PIELM, together with a dynamic weighting strategy. These components enable the model to effectively handle class imbalance in data streams and enhance its sensitivity to concept drift. Finally, an online ensemble learning framework is constructed with OS-PIELM as the base classifier to further improve the robustness of the proposed method. Extensive experiments on nine synthetic datasets and two real-world datasets demonstrate that the proposed method can effectively address class imbalance in data streams and improve concept drift detection performance.