| In the last decade,ensemble logic is more often applied to classification tasks,ensemble classification is a well-established approach that involves fusing the decisions of multiple predictive models.A similar “ensemble logic” has been recently applied to challenging feature selection tasks.At present,feature selection method based on ensemble learning is mainly divided into heterogeneous approaches and homogeneous approaches.In fact,regardless of heterogeneous approaches or homogenous approaches,the perturbation mode of the integration strategy is relatively single.Therefore,to solve blemish of existing ensemble feature selection algorithm,this paper proposes ensemble feature selection framework based on hybrid disturbance.This paper designs the ensemble feature selection method based on mixed disturbance by using function perturbation and data perturbation to improve the stability of the feature selection process and the predictive accuracy of the selected feature subset.Firstly,this paper uses Bootstrap to resample training samples to generate training data sets with different perturbation versions.Then,different feature selection algorithms are used to build integration components.Finally,the result of feature selection that different core algorithm is trained on different perturbed versions of the original data is combined into feature subset according to certain aggregation strategy.In order to ensure the realization of the algorithm,this paper designs a joint evaluation framework that can simultaneously calculate the predictive accuracy and stability.The validity of the method is verified by theoretical analysis and experimental evaluation.In addition,this paper explores the real impact of the integration implementation on the final selection outcome through a large number of similarity experiments.and validates that the influence of the ensemble components,thresholds and classifiers on the performance of the method through experiments with different ensemble configurations.Thus,some general suggestions are put forward for the design of the integration scheme.In summary,the main work of this article is a attempt for the ensemble feature selection.This paper achieves the purpose of simultaneously improve the predictive accuracy and stability and extends some wellknown studies that ensemble configuration.This paper is hoped to have a positive and effective meaning for the research of more comprehensive and effective digital ensemble feature selection in the future. |
