Dynamic Neural Network Ensemble Construction for Classification

Neural network ensemble (NNE) receives increasing attention in recent research among the e-commerce community. In an NNE method, multiple component neural networks are trained and cooperate with each other to solve the same problem. This paper presents a dynamic ensemble construction approach based on a coevolution paradigm. The whole model is obtained by a specially designed cooperative coevolutionary algorithm. After the coevolution process, a further heuristic structure refining process on the ensemble size is conducted in order to find the appropriate ensemble size for different datasets. The dynamic ensemble size value is obtained by removing less-contribution component networks. Experimental results illustrate that the classification performance of the proposed algorithm is superior to the traditional ensemble methods on real-world datasets.