Semi-supervised Online Elastic Stochastic Configuration Network to Deal with Concept Drift and Class Imbalance in Data Streams
Chaofan Chen, Wenhao Zhong, Kezhong Lu · 2024
Extreme learning machines (ELM) has been widely used in data stream processing, but because the parameters of ELM are generated randomly, the general approximation ability of randomized neural network depends on hidden layer nodes and random parameter range, if the parameters are improperly set, the objective function can not be approximated with high probability. Stochastic Configuration Network(SCN) solves this problem. In order to deal with the problem of data stream classification better, this paper proposes SSOE-FW-SCN, which combines the advantages of SCN and online elastic ELM. The algorithm adds class weights in the initialization phase and online update phase to adapt to the problem of class imbalance. The confusion matrix is used to dynamically calculate the class weight and forgetting factor, which makes the algorithm adapt to the problem of concept drift and class imbalance. In addition to parameter update, the algorithm also adds semi-supervised structure update in the online phase, which makes the algorithm adapt to more concept drift scenarios and class imbalance problems. Compared with the existing data stream classification algorithms, the proposed algorithm has better approximation and robustness.