Research And Application Of Selective Ensemble For Multiple Classifiers

Using machine learning to analyze the massive data in real life,the useful information obtained from it has a good guidance for people's production and life.Classification is one of the important tasks of machine learning.However,data in real life often have high latitude and noise,which seriously affects the accuracy and efficiency of classification.If we use the traditional base classifier to classify,it will be difficult to achieve good classification results,and the generalization ability of a single base classifier is not strong.Therefore,in order to effectively deal with the classification problems in real life,we need to consider building a good classifier and the combination of multiple base classifiers to form a new classifier is a good solution.In this paper,we aim to build a multi-classifier integration model,focusing on the method of integrating multi-base classifiers to form a strong classifier with excellent classification effect.This study has a certain reference value for improving the classification accuracy and generalization ability of classifiers.Based on the traditional base classifier,this paper focuses on multi-classifier integration model and multi-classifier selective ensemble model.The main contents and work are summarized as follows:1.According to the basic theory part,four aspects are discussed:1)Data preprocessing,which mainly aims at the disadvantage of classification due to the unbalanced distribution,redundancy and high-dimensional characteristics of data in daily life,and takes into account data normalization,data dimension reduction technology,unbalanced data processing methods,etc.2)Classification algorithm,analysis of traditional classification algorithms,including:Bayes,Support Vector Machine(SVM),Decision Tree,K Nearest Neighbor(KNN),Neural Network,Random Forest.The advantages and disadvantages of SVM and KNN are described in detail.3)Genetic algorithm,elaborates the important components of genetic algorithm and its implementation process.4)The evaluation criteria of classification algorithm are introduced,including accuracy,precision,recall rate,the meaning of F1 value and corresponding calculation formula.2.For the multi-classifier integration part,the Bagging algorithm,Boosting algorithm and Stacking algorithm are mainly analyzed.It is concluded that the classification effect of support vector machine in the base classifier is better than that of other base classifiers,so the support vector machine is selected as the base classifier,and the AdaBoost_SVM multi-classifier integration model and Bagging_SVM multi-classifier integration model are constructed.The experiments on breast cancer data set,hepatitis data set,bank marketing data set,earthquake data set and audit data set verify the performance of the two multi-classification integration models on classification problems.3.Aiming at the selective ensemble part of multi-classifier,the shortcomings of multi-classifier integration model,such as high time complexity and large space complexity,are analyzed.Therefore,a multi-classifier selective ensemble model is constructed.The implementation process of GMDH algorithm is given.A differential classifier is constructed based on GMDH algorithm,and the differential classifier is optimized by genetic algorithm.Finally,a multi-classifier selective ensemble model based on GMDH_GA is obtained.Finally,the performance of the algorithm is verified by the same data set.The experimental results show that the GMDH_GA multi-classifier selective ensemble model is superior to AdaBoost_SVM multi-classifier integration model,Bagging_SVM multi-classifier integration model and GMDH model in classification effect.