Meta-cognitive extreme learning machine for regression problems

In this paper, we present an efficient fast learning algorithm for regression problems using meta-cognitive extreme learning machine(McELM). The proposed algorithm has two components, namely the cognitive component and meta-cognitive component. The cognitive component is an extreme learning machine (ELM) while the meta-cognitive component which controls the cognitive component employs a self-regulating learning mechanism to decide what to learn, when to learn and how to learn. The meta-cognitive component chooses suitable learning method based on the samples presented namely, delete sample, reserve sample and network update. The use of ELM improves the network speed and reduces computational cost. Unlike traditional ELM, the number of hidden layers is not fixed priori in McELM, instead, the network is built during the learning phase. This algorithm is evaluated on a set of benchmark regression and approximation problems and also on a real-world wind force and moment coefficient prediction problem. Performance results in this study highlight that McELM can achieve better results compared with conventional ELM, support vector regression (SVR).