Online sequential extreme learning machine with forgetting mechanism

Abstract

The ensemble of online sequential extreme learning machine (EOS-ELM), an average of several online sequential extreme learning machines (OS-ELMs), can learn data one-by-one or chunk-by-chunk with fixed or varying chunk size. EOS-ELM provides higher accuracy with fewer training time, better generalization performance and stability than other popular sequential learning algorithms. However, in plenty of practical applications such as stock forecast, weather forecast, etc., training data often have timeliness, that is, each datum has a period of validity. In order to reflect the timeliness of training data in the process of learning, an improved EOS-ELM, called online sequential extreme learning machine with forgetting mechanism (FOS-ELM), is proposed in this paper. The proposed FOS-ELM cannot only retain the advantages of EOS-ELM, but also improve the learning effects by discarding the outdated data quickly in the process of learning to reduce their bad affection to the following learning. Detailed performance comparisons of FOS-ELM are carried out with EOS-ELM in the stock price short-term predictions. The experimental results show that FOS-ELM has higher accuracy with fewer training time, better stability and short-term predictability than EOS-ELM.