Research On The Method Of Classifier Selection Integration Based On Confusion Matrix

Integrated Learning is an important research direction in the field of machine learning,through training more individual classifiers and combine them to form multi-classification system,in order to improve the classification performance.At present,the integration technology has applied in many fields,such as handwriting recognition,satellite detection,face detection,text classification.Seen in this light,integrated learning has great room for development and broad application prospects.But with the development of computer science,the amount of data increases,the participation of an integrated classifier will more and more,so that on the one hand the rapid growth in the amount of calculation,on the other hand,the degree of difference between classifiers also becomes small,the impact of integrated accuracy,and an effective integrated system requires the participation of an integrated classifier has a relatively high accuracy and difference.Studies have shown that training produces base classifiers select part of an integrated,this method may use all over the base classifier to better integrate results.Therefore,the choice of having a high difference classifier from a large number of base classifiers as representatives integration has become a trend to integrate research study,the need for more in-depth study.Based on the integrated study,first introduced the multiple classifiers integration of domestic and foreign research background and significance,summarized the research status of integrated learning.Secondly,it introduces the concept of integrated learning and two classical ensemble algorithm Bagging and Boosting algorithm,then,were the product of the rules listed,six kinds of integration rules summation rules.Then measure from the perspective of the difference formula,we introduce the concept of diversity measure,as well as commonly used measure formula.Finally,a new multi-classifier selective method,the specific method is to construct all base classifiers confusion matrix as clustering of data objects,according to the distribution of the cluster sample and choose a certain number of classifiers as a representative,to be integrated to form a new collection of classifier,then this method is applied in thetraining process Bagging.In order to verify the feasibility of the proposed method,the UCI datasets experiments,the experimental results of the method of this article applies Bagging algorithm training process and the results obtained using raw Bagging algorithm were compared,show that this method can effectively improve the accuracy of the integrated system.And choose a different integration rules integration,analyzing the results.