Multiobjective Parameters Optimization of Extreme Learning Machine Based on MOEA/D

In recent years, Extreme learning machine (ELM) is widely used in data mining analysis and optimization problems. However, most existing ELMs are based on single-objective optimization, which may reduce the generalization performance of the model. To overcome this problem, a multiobjective parameter optimization strategy is proposed for ELM. This strategy simultaneously optimizes model error and generalization performance, which are two conflict objectives. Furthermore, the improved algorithm is called MOEA/D-ELM, which is used for supervised classification problems. The proposed algorithm in this paper has been tested on the UCI dataset. The experimental results show that MOEA/D-ELM is effective.