Gradient Boosting-Based Negative Correlation Learning

Negative correlation learning (NCL) is a useful ensemble learning approach, and has been used for neural network ensembles. In this paper, a new NCL algorithm, NCL. GBM is proposed, which uses gradient boosting machine (GBM) as the base learner. First, the feasibility of combining NCL and GBM is analysed. Then, we describe in detail how to apply negative correlation learning onto GMB. Empirical results show that NCL. GBM does have the ability to produce models with lower correlation and can usually result in lower generalization error than the original GBM.