Research About Integrated Classification Of Online Learning

In the field of the Internet, the data set is continuously over time to produce, so machine learning based on the complete data set is almost impossible. As it requires the learner can continue to learn, the traditional machine learning methods have been unable to meet this requirement. In view of the above situation, online learning came into being. At present the key techniques of online learning is incremental learning.On the other hand, SVM (Support Vector Machine) as an important research branch of machine learning, due to the perfect mathematical theory and the practical application effect, has many applications in many fields. However, the performance of support vector machine classifier is stable, so it can’t be used on incremental learning and it suffers from the catastrophic forgetting phenomenon. So how to get the support vector machine to be applied to the field of online learning has become a current research focus. Aiming at this problem, the article has done the following research:Firstly, a brief introduction to the mathematical theory, the basic concepts and the advantages of support vector machine.Then this paper gives a detailed description of a rapid classification algorithm of support vector machine. This classification algorithm based on the geometric structure information of sample set, starts with the selection of samples, which are located on the edge position of the sample set. Then it uses the samples which are hull vectors to form a new sample set for training support vector machines, in order to find out the optimal hyperplane. Since using a linear programming operation in the process of extracting the hull vectors and the new sample set of hull vectors accounted for only part of the original sample set, even a small part, the method can effectively reduce the complexity of the two planning operation process, so as to improve the speed of the algorithm. Secondly, in the process of extracting hull vector set, there is no loss of critical information, such as support vector, so there is no effect on the accuracy of the algorithm.Finally, based on the rapid classification algorithm, this paper proposes a new ensemble algorithm of support vector machines to support online learning——a Learn++ensemble method of support vector machines based on the hull vectors. This method introduces the hull vectors into the process of training support vector machine classifiers, and integrates the classifiers using Learn++algorithm. The integrated classifier can effectively perform incremental learning, but also reduce the training time and the storage size. The experimental results show that the integrated classifier not only has the ability of online learning, but also can avoid the phenomenon of catastrophic forgetting.