Sequential extreme learning machines for class imbalance and concept drift

Class imbalance and concept drift are two problems commonly exist in sequential learning. A weighted online sequential extreme learning machine (WOS-ELM) algorithm is proposed that has a distinctive feature of class imbalance learning (CIL) in both the chunk-by-chunk and one-by-one modes. A new sample can update the classifier without waiting for a chunk to be completed. For CIL in drifting environments, a computationally efficient framework, referred to as ensemble of subset online sequential extreme learning machine is proposed. It comprises a main ensemble representing short-term memory, an information storage module representing long-term memory and a change detector to promptly detect concept drifts. A self-regulatory method, referred to as meta-cognitive online sequential extreme learning machine, is proposed to adapt the learning according to the nature of data stream i.e. select appropriate strategy for class imbalance and concept drift learning. A single OS-ELM equation is proposed for multiclass imbalance and concept drift learning.DOCTOR OF PHILOSOPHY (EEE