Research On Support Vector Machine Classification Based On Selective Integration Learning

In recent years, with the rapid development of science and technology, leading to theincreasing of amount of data, it also brings a lot of problems, such as many data need to classifyand organize, pure rely on artificial classification of these data will produce a very largeworkload, and bring so much inconvenience to people’s work and life. And machine learning asa way to help people to solve this problem is getting more and more attention. Nowadays,machine learning in disease diagnosis and identification of biological information, geographicinformation systems, and other fields has a lot of application and research, and has brought a lotof convenience for people’s life and work. Ensemble learning as a kind of multiple classifierswhich can be integrated machine learning methods are playing an increasingly important role.Ensemble learning compared weak classifier individuals, its integration study effect is better.So the research on ensemble learning is focused on the combination of weak classifier, and theresearch on how to integrate the strong classifier with SVM is rare. To this, this paper’s mainwork can be divided into the following several aspects:1、 Put forward the selective integration of SVM based on Bagging algorithmCombined with Bagging algorithm and selective integration theory, the SVM based onBagging algorithm selective integration method is proposed. First, by the Bagging algorithm ofthe Bootstrap method, algorithm chosen the original data, get the training set, and thenrandomly selected feature subsets, projection on the training set and get the required input data,change the input samples through this way to solve the weakness of SVM classifier the problem,and makes some differences between each sub classifier; secondly, the algorithm combinedwith selective integration theory, for the precision on the sort of the classifier, pick out the partprecision better classifier for integration, to solve the integration of individual need to guaranteea certain accuracy to improve the overall performance problems, the overall performance isimproved, while reducing the computing resources.2、 Put forward the selective integration of SVM based on Adaboost algorithmCombined with the iterative weighting process of Adaboost algorithm and the selectiveensemble method, the selective ensemble method of SVM based on Adaboost is proposed.First, algorithm combining with the characteristics of the SVM classification, get the difficultand the easy part of the classification of data provided to the Adaboost algorithm, at the sametime in the process of iteration algorithm to adjust the nuclear parameters of SVM according tothe accuracy of calculation, through the two way to solve the problem of weakening the SVMindividual learning, at the same time to solve the individual learners difference, and theprediction accuracy to maintain at a certain level of theory; secondly, in the process of selectiveensemble method combined with integrated learning system, puts forward a new method of selecting child classifier, its by individual accuracy compared with the overall system accuracythat meet the requirements of the overall accuracy child classifier is added to the finalintegration learning system, improve the accuracy of the system, reduce the size of theintegrated system.The ensemble learning and strong learning type of classifier combination, the mainproblem is how to deal with the weakening learn, individual differences, and accuracy of thethree aspects of balance in the ensemble learning, only deal with these three aspects can beachieve better effect.