Research On Ensemble Learning Method Based On Diversity Weighted Evidece Reasoning Rule

As an important branch of machine learning,ensemble learning has been widely used in regression,classification,and clustering tasks by combining multiple base learners to obtain a model with better performance than a single base learner.Using average method and voting method as combination strategy in ensemble learning is difficult to effectively mine the output information of base classifiers,resulting in poor ensemble accuracy.The process of combining base classifiers needs to determine the weights of base classifiers,common weighting methods do not consider the influence of diversity on weight,which leads to the lack of reasonable weight assignment.In addition,the combination strategy integrates all samples using only one set of parameters,without paying attention to the adaptability of samples and parameters.In order to solve the above problems and further improve the performance of ensemble learning models,this paper conducts in-depth research on ensemble learning methods,the main content is as follows:(1)To fully mine the output information of the base classifiers and improve the rationality of setting weights.In this paper,the evidence reasoning(ER)rule is used as the combination strategy of base classifiers,and the diversity weighting method is used to set the weights of base classifiers,and the ensemble learning model based on diversity weighted evidence reasoning rule is constructed.The core content of this model is to calculate the difference values of base classifiers by diversity measure and normalize them into weights.After the verification of multiple data sets,the accuracy of the proposed method has been improved.(2)Evidence reasoning rule integrate all samples with a set of parameters,do not pay attention to the adaptability of samples and ER rule parameters,so that the accuracy of classification still has room for improvement.Considering the reliability and weight difference of base classifiers in different categories,this paper proposes a parameter adaptive ensemble learning model.According to the precision of the predicted results of the base classifiers,the model divided several sample clusters and calculated the parameters of ER rule for each sample cluster.Then it matched the test set samples with the sample clusters,and used the parameters of the corresponding sample clusters to integrate the samples through the ER rule.On the basis of ensuring that the combination strategy has good interpretability,the prediction accuracy of the ensemble learning model is further improved.(3)Implementation of prototype system for scene recognition based on ensemble learning.In order to popularize the research results of diversity weighting and parameters adaptive strategy in this paper,and enable users to quickly grasp the application method of the algorithms in this paper,a prototype scene recognition system is designed and implemented.The user can upload the scene image to be identified through the browser,and display the recognition results to the user through the feature extraction of the server and the integration of multiple deep learning models.In this paper,the research work on the weighting method of base classifiers and the adaptive parameters of ensemble learning can effectively improve the performance of ensemble learning model.The ensemble learning method based on the diversity weighted evidence reasoning rule proposed in this paper can effectively exploit the output information of the base classifiers,and the parameters are set to make the base classifiers achieve complementary effects,which improves the prediction accuracy.The precision of base classifiers is innovatively used to divide the ensemble granularity,and the adaptability of samples and combination strategy parameters is paid attention to when integrating multiple base classifiers,which further improves the recognition accuracy.The effectiveness of the proposed method is verified by experiments on multiple data sets.