Research On The Combination Of SVM And Adaboost Classification Algorithm

Support Vector Machines (SVM), proposed by Vapnik, is a machine learning technique based on the structural risk minimization principle, and it is also a new regression method with good generalization ability. SVM is broadly and greatly applied in pattern recognition, regression analysis and probability density, etc.AdaBoost, as one of the most famous boosting algorithms, has been used in different fields of machine learning. The approach allows designers to continually add new "weak classifiers", via assembly learning, until the learning approach obtains a predetermined error rate. Due to its prominent advantages, many people focus on the improvement of this algorithm in different ways.The thesis firstly presents a SVM parameters selection algorithm based on Fisher criteria-FS algorithm, which combines gradient descent algorithm with Fisher rules. The selection algorithm, combining with the gradient descent algorithm gets optimization parameter by making full use of the linear separable samples in the classes in the feature space. The selection algorithm has the advantages of simple, low complexity and easily implement etc. Secondly, the thesis studies on how to combine SVM into AdaBoost algorithm, and puts forward an efficient AdaBoost algorithm by combining SVM-IASVM algorithm, which also includes the nearest neighbor method and references to active learning strategies. Experimental results on a mid-size standard testing set that the two algorithms both can maintain83%above classification accuracy and the training time of IASVM is only one-tenth of the SVM’s. Therefore, the two algorithms greatly improve the classification performance, while maintain a good classification accuracy.