Feature Selection Using Stacking Integration and Partial Exploration Bayesian Optimization
To address the problems that the optimal feature subset of high-dimensional gene datasets is not easy to be determined and the traditional Bayesian optimization algorithm is prone to falling into local optimum,which cannot quickly select the optimal pa-rameters,in this paper,we propose a gene selection method based on the Stacking integration and partial exploration Bayesian opti-mization.Firstly,the Chi-square filtering scheme is used to eliminate the redundant genes in the original feature space,so as to ob-tain the genes with high correlation.The acquisition function of the Bayesian optimization algorithm is improved,and the jump out coefficient is introduced,so that the Bayesian optimization algorithm can adaptively jump out of the local optimum.The cost can be reduced and the efficiency of optimization will be speeded up.Secondly,the partial exploration Bayesian optimization is used to find the optimal parameters of random forest.Then,the optimized random forest model is employed to screen the optimal feature subset.Finally,a framework of the Stacking integration model is designed to construct classifier and classify the optimal feature subset,and then a gene selection algorithm based on the Stacking integration and partial exploration Bayesian optimization is constructed.The experimental results on nine public gene expression profile datasets show that the proposed algorithm can quickly select the optimal gene subset with higher classification accuracy.
gene selectionstacking algorithmbayesian optimization algorithmrandom forest model
国家科技期刊平台
NSTL
万方数据
