Handling Concept Drift in Time-Series Data: Meta-cognitive Recurrent Recursive-Kernel OS-ELM

Zongying Liu,Chu Kiong Loo,Kitsuchart Pasupa

DOI: https://doi.org/10.1007/978-3-030-04224-0_1

2018-01-01

Abstract:This paper proposes a meta-cognitive recurrent multi-step-prediction model called Meta-cognitive Recurrent Recursive Kernel Online Sequential Extreme Learning Machine with a new modified Drift Detector Mechanism (Meta-RRKOS-ELM-DDM). This model combines the strengths of Recurrent Kernel Online Sequential Extreme Learning Machine (RKOS-ELM) with the recursive kernel method and a new meta-cognitive learning strategy. We apply Drift Detector Mechanism to solve concept drift problem. Recursive kernel method successfully replaces the normal kernel method in RKOS-ELM and generates a fixed reservoir with optimised information. The new meta-cognitive learning strategy can reduce the computational complexity. The experimental results show that Meta-RRKOS-ELM-DDM has a superior prediction ability in different predicting horizons than the others.