Research On A Combined Classification Based On Algorithm Based On AdaBoost

Classification is an important kind of research method to be used for solving the class label decision problems in data mining field. It is also an important technology for data analysis at the same time. As an effective kind of tool in scientific study, combination classifiers have been applied widely in the fields such as the medical diagnosis, handwriting recognition, face recognition and recommendation system etc.In this thesis, the basic theory of classification and the classic algorithms of DM theory are both analyzed at first. The reasons of the advantages of ensemble model are discussed by comparing with the single model. And then, the basic principles and algorithms of ensemble(model) learning are discussed in detail. For Bagging and Boosting algorithms, contrastive studies have been investigated both in theory analysis and experimental verification.Based on researched basis, the algorithms and their improvements have been analyzed from the algorithms diversity point of view of training or learning under the idea of collective wisdom and ensemble learning theory, and a new framework method or algorithm named Ensemble-Ada Boost is put forward. In this method, the base classifiers or basic classification algorithms are different and can be applied to several kinds of algorithm frameworks to form several kinds of base classifier. At last, the classification outputs of these base classifiers are combined in the way of simple majority voting which are used as the results output by the Ensemble-Ada Boost algorithm. By increasing the diversities of base classifiers, as a whole, the classification error of the new ensemble classifier or algorithm Ensemble-Ada Boost is effectively reduced, and the generalization ability and the stability of this algorithm are both improved also.