Research On Dynamic Ensemble Classification Method

Dynamic ensemble is not only one of the key research directions of machine learning,but also widely used in many fields.Although dynamic ensemble technology has been widely developed,with the advancement of technology,people have higher and higher requirements for the performance of integrated systems.Therefore,how to improve the generalization performance of the system and improve the classification accuracy is still one of the problems to be solved.The research of dynamic ensemble methods not only has far-reaching theoretical significance,but also has wide application value.Based on the above research background,this paper focuses on the optimization of dynamic ensemble algorithms,and proposes corresponding solutions.The main research contents of this paper include the following aspects:(1)This paper proposes a margin-based heterogeneous classifier generation method.Based on the analysis of the theory of the margin,the method introduces the concept of the margin into the training of the base classifier,and guides the training of the individual classifier by calculating the margin value of the training sample.This method not only finds the most accurate set of base classifiers in the iterative loop,but also increases the difference between classifiers based on the heterogeneous base learning algorithm.After experimental comparison,it is verified that the method can improve the accuracy of the ensemble learning algorithm.(2)In this paper,the dynamic ensemble selection algorithm is sensitive to the dynamic selection data set distribution,especially when the dynamic selection data set contains noise,highly overlapping sample instances,the classification effect is not ideal,and the capability area cannot fully represent the sample to be tested.An improved dynamic ensemble selection algorithm is proposed for the edited nearest neighbor and adaptive nearest neighbor selection mechanisms.In this paper,the sensitivity of the algorithm is reduced by filtering out the negative samples in the dynamic selection dataset,and the most representative capability region is selected by the adaptive distance method,so that the classification ability of the base classifier is reasonable measured in the optimal capability region,to achieve the goal of improving system accuracy.Through experiments,the method reduces the running time,improves the classification accuracy,and enhances the stability of the integrated system.