An online learning neural network ensembles with random weights for regression of sequential data stream.

An ensemble of neural networks has been proved to be an effective machine learning framework. However, very limited studies in the current literature examined the neural network ensemble for online regression; furthermore, these methods were combination of online individual models and did not consider the ensemble diversity. In this paper, a novel online sequential learning algorithm for neural network ensembles for online regression is proposed. The algorithm is built upon the decorrelated neural network ensembles (DNNE) and thus referred to as Online-DNNE; so it uses single-hidden layer feed-forward neural networks with random hidden nodes' parameters as ensemble components and introduces negative correlation learning to train base models simultaneously in a cooperative manner which can effectively maintain the ensemble diversity. The Online-DNNE only learns the newly arrived data, and the computation complexity is thus reduced. The results of the experiments with benchmarks show the effectiveness and significant advantages of the proposed approach. [ABSTRACT FROM AUTHOR]