Selective Hierarchical Ensemble Modeling Approach and Its Application in Leaching Process

Abstract

To improve the precision and generalization of ensemble model and leaching model, a novel selective hierarchical ensemble modeling approach is proposed for leaching rate prediction in this paper. Unlike previous selective ensemble model, the new selective ensemble model is a hierarchical model. The model considers not only the combination of sub-models, but also the generation of sub-models. First of all, a new multi-model ensemble hybrid model (MEHM) based on bagging algorithm is proposed. In this model, the sub-models are composed of data model and mechanism model. The data model generates training subsets by using the proposed based vector bootstrap sampling algorithm. Afterwards, a new selective multi-model ensemble hybrid model (NSMEHM) based on binary particle swarm optimization (PSO) algorithm is presented. In this model, the binary PSO optimization algorithm is used to find out a group of the MEHMs, which minimizes the error and maximizes the diversity. Experiment results indicate that the proposed NSMEHM has better prediction performance than the other models.