Reasearch Of Selective Ensemble Learning And Its Appliacation

Ensemble learning is a machine learning paradigm.Compared with the traditional single classifier,ensemble learning can significantly improve the classifier generalization ability by combining base classifiers.Currently,ensemble learning has been applied widely in the field of disease diagnosis,bio-metric recognition,text classification and information processing.As a new direction of ensemble learning,selective ensemble learning trying to exclude some classifiers that has poor performance to get better ensemble ability.However,how to select the base classifiers and how to decide the optimal combination of classifiers still need further study in the selective ensemble system.Research of this paper has important theoretical and practical significance on improving the performance of ensemble learning and promoting ensemble learning in practice.This paper studies the basic theory of ensemble learning,ensemble learning approach,selective ensemble learning and its applications,focuses on the study of base classifier selection,The main research content and results of the work are as follows:1.We have made in-depth analysis of ensemble learning theory,and have a research on classifier generator,classifier fusion methods and the evaluation of ensemble performance.Taking into account of the complexity in the data application,we research the data normalization,dimension reduction and unbalanced data processing methods,which will lay the foundation for future study.2.Ensemble learning can significantly improve the generalization ability of a learning system,and has received wide attention from the machine learning community.Combining the Bagging algorithm and decision tree in machine learning methods,we establish a diagnostic model to analysis the medical disease data.Comparing the performance of ensemble learning model with a single decision tree diagnostic model,experimental results verify that the ensemble learning method has better diagnostic accuracy.3.The accuracy and diversity of classifiers plays a vital role in ensemble learning machine.For the base classifier selection problem,we propose a BAD(Balancing Accuracy and Diversity)selection rules;On this basis,we establish a selective ensemble model of BAD and make a experiment on the different proportion of accuracy and diversity,and analyzing which proportion is best for BAD,experimental results prove the validity of BAD guidelines.On the base classifier selection method,we proposed a new FBGA(Forward and Backward Genetic Algorithm)selection method using the genetic algorithm combined with the preamble and subsequent selection ideas,experimental results show that FBGA method have a better performance in the search results.